A method, device and equipment for predicting game results based on machine learning

Abstract

The present invention relates to a method, device and system for predicting game results based on machine learning, comprising: determining contestants of a target game; determining feature data of the contestants based on historical game data of the contestants; obtaining game result interference data of the contestants; predicting the game result of the feature data and the game result interference data based on a first game result prediction model to obtain first game result data of the target game, and improving the accuracy of game result prediction by learning the feature data and the game result interference data of the contestants through the game result prediction model.

Description

Technical Field

[0001] The present disclosure relates to the technical field of machine learning, and in particular to a method, device and apparatus for predicting game results based on machine learning. Background Art

[0002] At present, many methods have emerged to predict the results of competitive games based on strict mathematical statistics. According to the information we have, there are three authoritative prediction methods that are currently widely used to predict the results of football matches:

[0003] 1. ELO prediction method

[0004] The Elo prediction method was created by physicist Dr. Aroad Elo. Dr. Elo first used this method to predict the results of chess games. He explained the method in detail in his book "Chess: Past and Present". Through the study of 1,500 Premier League games, Jacques Black has made unremitting improvements to the Elo prediction method, which has now been widely used in football matches. The improved model of the Elo prediction method predicts the outcome by studying the points of the home and away teams before the game. The prediction regression equation of the Elo prediction method is:

[0005] The probability of the home team winning = 44.8% + (0.53% multiplied by the points difference between the two teams)

[0006] The away team's chance of winning = 24.5% + (the points difference between the two teams multiplied by 0.39%)

[0007] 2. Goal Rate Comparison Prediction Method

[0008] In 1990, David Jackson and K. R. Mosheski published a paper at the International Gambling Conference, "Index Gambling in Games", which proposed for the first time a mathematical model that uses the average goal rate per game as a prediction of a team's next game performance. This method has the highest accuracy in predicting the results of the National O Super League and the National L First League.

[0009] Simply put: There are four principles in the goal rate prediction method:

[0010] 1. When the difference between the average goal rate of the two teams is 0.30 (excluding 0.30) or more, the team with the higher average goal rate wins;

[0011] 2. When the difference between the average goal rate of the two teams is 0.10 or more and 0.30 (inclusive), if the home team has a higher average goal rate, the home team wins;

[0012] 3. When the difference between the average goal rate of the two participating teams is between 0.10 and 0.30 (inclusive), if the average goal rate of the home team is lower than that of the away team, the home team wins or draws.

[0013] 4. When the difference between the average goal rate of the two participating teams is 0.10 (inclusive) or less, the home team wins or draws.

[0014] 3. Comparison prediction method of the last six games

[0015] This prediction method was first provided to lottery players by newspapers in Country X. It has four main principles:

[0016] 1. When the points difference between the two teams in the last six games is 6 or more, the team with the higher points in the last six games wins;

[0017] 2. When the difference in points between the two teams in the last six games is 5, if the home team has a higher score in the last six games, the home team wins; if the home team has a lower score in the last six games, the home team wins or draws;

[0018] 3. When the difference in points between the two teams in the last six games is 2 to 4, the team with the higher points in the last six games wins;

[0019] 4. When the points difference between the two teams in the last six games is 1 or 0, the home team wins or draws.

[0020] The above three traditional methods mainly use simple statistical principles to fit and predict football results. The model's generalization ability is seriously insufficient, and it is impossible to evaluate the impact of sudden random events on the results (red and yellow cards, injuries, player conflicts, etc.). Due to the inherent confrontational nature of football, 90 minutes is a collection of "random processes", and the prediction accuracy of the above models in a long time window has no practical application value. Therefore, a new method for predicting game results needs to be proposed. Summary of the invention

[0021] The present disclosure proposes a method, device and equipment for predicting game results based on machine learning, which can greatly improve the security of user wireless networks. The present disclosure is specifically implemented by the following technical solutions:

[0022] In one aspect, the present disclosure provides a method for predicting game results based on machine learning, comprising:

[0023] Determine the participants for the target competition;

[0024] Determining characteristic data of the contestant based on historical competition data of the contestant;

[0025] Obtaining game result interference data of the contestant;

[0026] The match result is predicted based on the first match result prediction model for the feature data and the match result interference data to obtain first match result data for the target match.

[0027] On the other hand, the present disclosure provides a game result prediction device based on machine learning, comprising:

[0028] A first determination module is used to determine the participants of the target competition;

[0029] A second determination module, configured to determine feature data of the contestant based on historical competition data of the contestant;

[0030] An acquisition module, used to acquire the game result interference data of the contestant;

[0031] The first prediction module is used to predict the match result of the feature data and the match result interference data based on a first match result prediction model to obtain first match result data of the target match.

[0032] In some embodiments, the second determining module includes:

[0033] A second determining unit, configured to determine the confrontation characteristic data of the team contestant based on the win-loss information in the historical match data of the team contestant, wherein the historical match data includes match data of matches within a preset time period;

[0034] A third determining unit, configured to determine the competition status data of the team competition object based on the historical competition data;

[0035] A second acquisition unit is used to acquire first characteristic data of a contestant in the team contestant object;

[0036] A third acquisition unit is used to acquire second characteristic data of the strategic instructor among the team contestants;

[0037] The first processing unit is used to use the confrontation feature data of the team contestant, the competition status data of the team contestant, the first feature data of the contestant and the second feature data of the strategic instructor as the feature data of the team contestant.

[0038] In some embodiments, the second determining module includes:

[0039] A fourth acquisition unit, configured to acquire score information of the game of the team contestants within the preset time period from the historical game data;

[0040] A first calculation unit is used to calculate the number of goals conceded and the entropy of goals conceded in the game of the team participating object within a preset time period according to the score information;

[0041] A fifth obtaining unit, configured to obtain, from the historical game data, ranking information of the team's competitors in the games within the preset time period;

[0042] A second calculation unit, configured to calculate ranking growth rate information of the team contestants according to the ranking information;

[0043] A sixth obtaining unit, configured to obtain the competition opponents of the team competition object from the historical competition data;

[0044] A fourth determining unit, configured to determine the strength information of the opponent;

[0045] The fifth determining unit is used to determine the competition status data of the team competition object according to the number of goals conceded, goal conceded entropy, ranking information and opponent strength information of the game within the preset time period.

[0046] In some embodiments, the second determining module includes:

[0047] A sixth obtaining unit, configured to obtain score information of a second number of recent games of the team contestants from the historical game data;

[0048] A third calculation unit is used to calculate the score standard deviation and score entropy of the team contestants in the second number of recent games according to the score information;

[0049] A seventh obtaining unit, configured to obtain, from the historical game data, information on the number of mistakes made by the team participants in a second number of recent games;

[0050] A fourth calculation unit, configured to calculate the error standard deviation and error entropy of the team contestant in the second most recent number of games according to the error number information;

[0051] A sixth determining unit, configured to determine, from the historical game data, game status data corresponding to a third number of recent games of the first preset game type;

[0052] A seventh determining unit, configured to determine the winning number information of the preset event type game according to the game status data corresponding to the preset event type game;

[0053] An eighth determination unit is used to determine the competition status data of the team participant based on the score standard deviation and score entropy, error standard deviation and error entropy of the team participant's most recent second number of games and the number of wins of games of a preset event type.

[0054] In some embodiments, the second determining module includes:

[0055] A ninth determining unit, configured to determine, based on the historical competition data, ranking information and points information of the participant in the event corresponding to the target competition;

[0056] A tenth determining unit, configured to determine game status data corresponding to a game of a second preset game type from the historical game data;

[0057] An eleventh determining unit is used to determine the competition status data of the team participant based on the ranking information, the points information and the competition status data.

[0058] In some embodiments, the second determining module includes:

[0059] An eighth obtaining unit, configured to obtain the game data of the team participants in the preliminary round from the game data of the preliminary round;

[0060] The second processing unit is used to use the game status data of the team contestant in the preliminary game as the competition status data of the team contestant.

[0061] In some embodiments, the second determining module includes:

[0062] a twelfth determining unit, configured to determine the confrontation characteristic data of the individual contestant based on the win-loss information in the historical match data of the individual contestant, wherein the historical match data includes match data of matches within a preset time period;

[0063] a thirteenth determining unit, configured to determine the competition status data of the individual competition object based on the historical competition data;

[0064] A ninth acquisition unit, configured to acquire third characteristic data of the individual contestant;

[0065] The third processing unit is used to use the confrontation feature data of the individual contestant, the competition status data of the individual contestant and the third feature data of the contestant as the feature data of the individual contestant.

[0066] In some embodiments, the acquisition module includes:

[0067] a tenth obtaining unit, configured to obtain information on incidental events between the contestants and between the contestants;

[0068] A fourteenth determining unit is used to determine the game result interference data based on the incidental event information.

[0069] On the other hand, the present disclosure provides a match result prediction device based on machine learning, the device comprising a processor and a memory, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the match result prediction method based on machine learning as described in the first aspect.

[0070] The present disclosure provides a method, device and equipment for predicting game results based on machine learning. By determining the contestants who need to participate in the target game, the contestant feature data determined based on the contestant's historical game data and the contestant's game result interference data are used as test data for predicting the game result. The test data is tested in combination with the game result prediction model to obtain accurate game results. The game result prediction model based on machine learning is introduced to predict the game result, replacing the method of predicting the game result using only simple statistical principles in the prior art. Machine learning is performed in combination with the two factors of the contestant's feature data and the game result interference data, thereby improving the accuracy of the game result prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0072] Figure 1 It is a schematic diagram of a two-layer classification model prediction process of a match result prediction method based on machine learning provided by an embodiment of the present invention;

[0073] Figure 2 It is a flowchart of a method for predicting game results based on machine learning provided by an embodiment of the present invention;

[0074] Figure 3 It is a schematic diagram of a process flow of a method for determining characteristic data of a contestant provided by an embodiment of the present invention;

[0075] Figure 4 It is a flowchart of a method for determining competition status data of a competition object provided by an embodiment of the present invention;

[0076] Figure 5 It is a flowchart of another method for determining the competition status data of a competition object provided by an embodiment of the present invention;

[0077] Figure 6It is a flowchart of another method for determining the competition status data of a competition object provided by an embodiment of the present invention;

[0078] Figure 7 It is a flowchart of another method for determining the competition status data of a competition object provided by an embodiment of the present invention;

[0079] Figure 8 It is a schematic diagram of another method for determining characteristic data of a contestant provided by an embodiment of the present invention;

[0080] Fig. 9 It is a flow chart of a method for determining interference data of a contestant's game result provided by an embodiment of the present invention;

[0081] Fig.10 It is a structural schematic diagram of a game result prediction device based on machine learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0083] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0084] The following describes the method for predicting game results based on machine learning of the present invention. This specification provides method operation steps as described in the embodiments or flow charts, but more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the actual system or server product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, in a parallel processor or multi-threaded processing environment).

[0085] Figure 1 FIG. 1 is a schematic diagram of a two-layer classification model prediction process of a method for predicting game results based on machine learning provided by an embodiment of the present invention. Figure 1 As shown in Figure 1, the prediction process includes:

[0086] Input the training samples with win, draw and loss results into the XGBoost model for training to obtain the XGBoost three-classification model;

[0087] Input the training samples with the winning and losing game results into the support vector machine (SVM) model for training to obtain the SVM binary classification model;

[0088] Then input the test samples of the actual predicted matches into the XGBoost three-classification model to obtain the prediction results of win, draw and loss;

[0089] The prediction result of the tie in step 3 is input into the SVM binary classification model to obtain the prediction result of winning or losing.

[0090] The present disclosure can not only predict the results of various competitive games, such as sports games, talent shows, cultural games, online games, etc., but also predict the results of games participated by individual contestants and group contestants.

[0091] Figure 2 FIG. 1 is a flow chart of a method for predicting game results based on machine learning provided by an embodiment of the present invention. Figure 2 As shown, the method specifically includes:

[0092] S201: Determine the participants of the target competition.

[0093] In the embodiments of this specification, the target game is a game that needs to be predicted based on the game results of machine learning. It can be a game that already has game results (used to confirm the game results in the experimental stage) or a game that will be held in the future (used to predict the game results).

[0094] In this step, the participants that need to be determined are the participants of both sides of the game, which can be individual participants or team participants. Team participants can be multiple players or one or more players and a strategic instructor (for example, a coach, mentor, etc.).

[0095] S203: Determine feature data of the contestant based on the historical competition data of the contestant.

[0096] In the embodiments of this specification, the characteristic data of the contestants are data that can reflect the objective strength of the contestants in the competition, for example, in sports competitions, the physical fitness of the contestants and the level of teamwork of the contestants; in dance competitions, the physical coordination ability of the contestants and their ability to understand and express music, etc. Competition result interference data are data other than the characteristic data of the contestants that can affect the result of the competition.

[0097] S205: Obtaining game result interference data of the contestant.

[0098] The obtaining of the game result interference data of the contestants includes: obtaining incidental event information between the contestants and the contestants; and determining the game result interference data based on the incidental event information.

[0099] S207: Predicting a match result of the feature data and the match result interference data based on a first match result prediction model to obtain first match result data of the target match.

[0100] The first match result prediction model is established in the following way:

[0101] Obtaining historical match data of a training contestant with a match result of win, draw or loss, wherein the training contestant includes the contestant;

[0102] Determining characteristic data of the training contestant based on historical match data with match results of win, draw or loss;

[0103] Obtaining game result interference data of the training contestant;

[0104] The preset machine learning model is trained based on the characteristic data of the training contestants and the game result interference data to obtain a first game result prediction model.

[0105] When the first match result data is a tie, the method further includes:

[0106] S209: Predicting a match result based on the feature data of the contestant and the match result interference data of the contestant, to obtain second match result data of the target match;

[0107] The second match result prediction model is established in the following way:

[0108] Acquire historical competition data of a training contestant with a win or loss result, wherein the training contestant includes the contestant;

[0109] Determining characteristic data of the training contestant based on historical competition data with win or loss results;

[0110] Obtaining game result interference data of the training contestant;

[0111] The preset machine learning model is trained based on the characteristic data of the training contestants and the interference data of the game results to obtain a second game result prediction model.

[0112] Regarding step S203, when the contestant is a team contestant, Figure 3 As shown, the determining of the characteristic data of the contestant based on the historical competition data of the contestant includes:

[0113] S301: Acquire first characteristic data of the team members in the team competition object.

[0114] Specifically, the first characteristic data is data that can reflect the strength of the contestants themselves, and may include age, the number of competitions participated in, the level of competitions participated in, etc. For example, in a singing competition, singers aged 18-23 may not have much experience in competitions and time to practice singing, while singers aged 24-20 may have more experience in competitions and time to practice singing, and thus their strength will be relatively stronger than singers aged 18-23.

[0115] S303: Acquire second characteristic data of the strategic advisor among the team participants.

[0116] Specifically, the second characteristic data is data that can reflect the strength of the strategic instructor, which may include the level of the competitions that have been coached, how long the experience of coaching related competitions is, etc. For example, in table tennis competitions, the coach who has coached the national team is generally stronger than the coach of the provincial team, and the coach who has served as a coach for five years is stronger than the coach who has served for one year.

[0117] S305: Determine the confrontation feature data of the team contestant based on the win-loss information in the historical match data of the team contestant, wherein the historical match data includes match data of matches within a preset time period.

[0118] The preset time period may be the most recent preset number of years, such as the most recent 4 years.

[0119] Correspondingly, the determination of the confrontation feature data of the team participants based on the victory and defeat information in the historical game data of the team participants is to determine the corresponding confrontation feature data based on the victory and defeat information of the games played against each other by the two team participants within the recent preset number of years.

[0120] S307: Determine the competition status data of the team participants based on the historical competition data.

[0121] Specifically, the competition status data includes fluctuating status data and fixed status data. The fluctuating status data reflects the ups and downs of the team's participating objects in the historical competition process, and the fixed status data reflects the static status of the current team's participating objects in the corresponding competition.

[0122] Fluctuation state data can be data within a certain period of time. Both individual contestants and team contestants can use the following method to determine the competition state data. Figure 4 As shown, the determining of the competition status data of the team competition object based on the historical competition data includes:

[0123] S401: Obtaining score information of the game of the team participants within the preset time period from the historical game data.

[0124] Those skilled in the art can understand that if one side scores in a game, it means that the other side loses points.

[0125] S403: Calculate the number of goals conceded and the number of goals conceded of the team in a preset time period according to the score information.

[0126] Specifically, in a football match, the number of goals conceded corresponds to the number of goals conceded. In addition, the goal conceded entropy is used to measure the uncertainty of the number of goals conceded, and the entropy in the goal conceded entropy can represent the Shannon entropy.

[0127] S405: Obtaining ranking information of the team participants in the competition within the preset time period from the historical competition data.

[0128] S407: Calculate ranking growth rate information of the team participants according to the ranking information.

[0129] S409: Acquire the opponents of the team participants from the historical game data.

[0130] S411: Determine the strength information of the opponent.

[0131] S413: Determine the competition status data of the team participants according to the number of goals conceded, goal conceded entropy, ranking information and opponent strength information of the game within the preset time period.

[0132] Fluctuation status data can be data from a certain number of games. Both individual and team participants can use the following method to determine the competition status data. Figure 5 As shown, the determining of the competition status data of the team competition object based on the historical competition data includes:

[0133] S501: Obtaining score information of a second number of recent games of the team participants from the historical game data.

[0134] The second most recent number of fields may be the most recent six fields.

[0135] S503: Calculate the score standard deviation and score entropy of the team's most recent second number of games according to the score information.

[0136] S505: Obtaining information on the number of mistakes made by the team participants in the second most recent number of games from the historical game data.

[0137] S507: Calculate the error standard deviation and error entropy of the team's most recent second number of games according to the error number information.

[0138] S509: Determine the game status data corresponding to the third number of games of the first preset event type most recently from the historical game data.

[0139] The number of the third number fields may be the same as or different from the number of the second number fields.

[0140] The first preset event type of competition may be a core competition, which may be determined based on the event organizer, the nature and importance of the competition, or may be other specific types of competitions that can reflect the ups and downs of the contestants' status.

[0141] S511: Determine the winning number information of the preset event type game according to the game status data corresponding to the preset event type game.

[0142] S513: Determine the competition status data of the team participant based on the score standard deviation and score entropy, error standard deviation and error entropy of the team participant's most recent second number of games and the number of wins of games of a preset event type.

[0143] Both individual and team participants can use the following method to determine the status of the competition. Figure 6 As shown, the determining of the competition status data (fixed status data) of the team competition object based on the historical competition data includes:

[0144] S601: Determine the ranking information and points information of the participant in the event corresponding to the target game based on the historical game data.

[0145] S603: Determine the game status data corresponding to the game of the second preset game type from the historical game data.

[0146] The second preset event type may be the same as or different from the first preset event type.

[0147] S605: Determine the competition status data of the team participants based on the ranking information, the points information and the competition status data.

[0148] Fixed status data may include data of the preliminaries of the target competition. Both individual participants and team participants may determine the competition status data using the following method. Figure 7 As shown, when the historical game data includes game data of a preliminaries of the target game, determining the competition status data of the team competition object based on the historical game data includes:

[0149] S701: Obtaining the game data of the team participants in the preliminary round from the game data of the preliminary round.

[0150] S703: Using the competition status data of the team contestant in the preliminary round as the competition status data of the team contestant.

[0151] The match status information of the target competition's preliminaries is also important data that can reflect the current status of the participants.

[0152] In the embodiments of the present specification, for obtaining the data of the competition status, the final competition status data can be obtained by taking a weighted average of the fluctuation state data and the fixed state data after respectively obtaining the fluctuation state data and the fixed state data, or it can be set according to actual needs.

[0153] In the embodiment of this specification, the competition status data can be based on the following Figure 4-7 The method described can also be based on Figure 4-7 One or more of the methods described are combined to obtain.

[0154] S309: taking the first characteristic data of the player, the second characteristic data of the strategic instructor, the confrontation characteristic data of the team participant and the competition status data of the team participant as the characteristic data of the team participant.

[0155] When an individual is a participant, the characteristic data of the individual participant is similar to the characteristic data of the team. Figure 8 As shown, when the contestant is an individual contestant, the feature data of the contestant is determined including:

[0156] S801: Acquire the third characteristic data of the individual contestant.

[0157] The third characteristic data of the individual contestant is consistent with the first characteristic data of the contestants in the team contestant, except that there are one or more contestants in the team contestant.

[0158] S803: Determine the confrontation feature data of the individual contestant based on the win / loss information in the historical match data of the individual contestant, where the historical match data includes match data of matches within a preset time period.

[0159] S805: Determine the competition status data of the individual participant based on the historical competition data.

[0160] S807: Using the third characteristic data of the contestant, the confrontation characteristic data of the individual contestant, and the competition status data of the individual contestant as characteristic data of the individual contestant.

[0161] Regarding step S205, obtaining the game result interference data of the contestant specifically includes: Fig. 9 As shown:

[0162] S901: Obtain conflicting information of the contestant.

[0163] S903: Obtain the injury information of the participant.

[0164] S905: Acquire media event information of the contestant.

[0165] S907: Obtain the ban information of the contestant.

[0166] S909: Determine the game result interference data based on the conflict information, injury information, media event information and suspension information.

[0167] The contradictory information, injury information, media event information and suspension information are all non-strength information about the contestants, but may have a significant impact on the results of the game. Using game result interference data as a factor in determining the game results can greatly improve the accuracy of game result prediction based on machine learning.

[0168] Based on the above description of the match result prediction method based on machine learning, the following continues to use a football match in a sports game as an example to further describe the process details of the match result prediction method based on machine learning.

[0169] Step 1: Crawling and entering team data

[0170] Through the crawler, the team lineup data, historical performance data, team ranking, historical red and yellow card data, and coach information data are collected. Taking the 2016 European Cup as an example, the present invention integrates the top 24 data to generate 6 basic dimension data tables (including all team historical confrontation records, coach information, past confrontation records between the top 24, basic information of the top 24, team historical rankings, and red and yellow cards of the team since the qualifiers).

[0171] Step 2: Development of Feature Wide Table

[0172] The characterization of team characteristics is mainly carried out in the following aspects: strength part and black swan part. Among them, the strength part is further divided into four parts: "team", "coach", "player", and "historical confrontation". In addition to the preprocessing of basic data, each part also includes deep processing with statistical and physical significance - the development of nested features, such as the goal / loss entropy and ranking points ratio that characterize the ups and downs of the team's status. There are a total of 5 major dimensions, including team dimension, coach dimension, player dimension, historical confrontation dimension, and black swan event, and 41 sub-dimensional features.

[0173] Specifically, the team dimension includes: whether it is a strong team (according to FIFA ranking), 1 for yes, 0 for no; points in the qualifying round; goal difference in the qualifying round; number of goals conceded in the qualifying round; world ranking; ranking points ratio; whether it has a top 5 record in three Class A competitions, 1 for yes, 0 for no; state fluctuations (characterizing dark horses, with the timeline extended beyond the qualifiers) sub-dimensions include: state value (developed based on the fitted linear empirical formula); number of wins in Class A competitions in the past six years; number of goal differences in the past six games in the past six years (positive, negative, 0); standard deviation of goal differences in the past six games in the past six years; standard deviation of goals conceded in the past six games in the past six years; entropy of goal difference / entropy of goals conceded in the past six games in the past six years; number of goals conceded in a single game (positive, negative, 0); entropy of goals conceded in a single game; whether the weak has defeated the strong, 1 for yes, 0 for no; world ranking growth rate from 2012 to 2016.

[0174] Specifically, the coaching dimensions include: whether the coach has a top 5 record in three Class A competitions, 1 for yes, 0 for no; whether the coach has coached the current team for more than 3 years, 1 for yes, 0 for no; whether the coach has coached a world-class team, 1 for yes, 0 for no; and the winning rate of the team.

[0175] Specifically, the player dimension includes: star distribution, player age distribution, and player experience distribution sub-dimensions. Among them, the star distribution sub-dimension includes: number of stars; percentage of stars; percentage of super stars; number and percentage of striker stars; number and percentage of midfield stars; number and percentage of defensive stars; the player age distribution sub-dimension includes: percentage of people aged 18-23; percentage of people aged 24-29; percentage of people aged 29 and above; the player experience distribution sub-dimension includes: percentage of people participating in the European Cup for the first time; percentage of people participating in the European Cup multiple times; percentage of people participating in 1-5 Class A games in the past five years; percentage of people participating in more than 5 Class A games in the past five years.

[0176] Specifically, the historical confrontation dimensions include: the number of wins and losses in the last three Class A matches; the number of wins in historical confrontations; and the goal difference in historical confrontations.

[0177] Specifically, the black swan dimensions include: media events, good 1, bad 0; conflicts between players and head coaches, yes 1, no 0; injury rate; whether there are red and yellow card suspensions, yes 1, no 0.

[0178] Among them, the quantification of the team status value is a key indicator, and we adopted the following derivation process:

[0179] Since the opponents of each team in the six selected games are different, in order to eliminate the "unfairness" caused by the difference in the strength of the opponents and make the weights of depicting historical confrontations more reasonable, we refer to the empirical formula formulated by FIFA - "Strength of opponent ST (Strength of opponent)", and its value is obtained according to the rules of FIFA for calculating the world rankings of each team: ST = (200-OR) / 100, where OR is the opponent's world ranking.

[0180] The actual score F'n is expressed by the nominal scores of the six games (n=1, 2...6) and the product of Fn and the strength of the opponent, that is, F'n=Fn*STn.

[0181] The actual score per game is .

[0182] In order to make the weight of historical confrontations more reasonable, we introduce the "status value" Z. For the 24 participating teams, each has its actual average score per game. We find the largest value , and then use their own than Get a value between [0, 1], this data is the "status value" that can accurately quantify the team's performance: Z = / .

[0183] The "status value" falls within the interval [0, 1]. The higher the "status value", the better the team's performance in the World Cup and the more likely it is to achieve good results; the lower the "status value", the less likely it is that the team will perform at its normal level in the game. The calculation results of the status value are shown in the following table. As the original data is updated, the status value is also updated.

[0184]

[0185] Table 1 Example of state value calculation results

[0186] The black swan part is the most attractive, gimmicky and unexpected feature of football. Due to the preparation time of the game, we mainly discretize the features for statistics, including red and yellow cards, injuries, conflicts between players and coaches, media gimmicks, etc.

[0187] Step 3: Use XGBOOST for three-category (win, lose, draw) training and prediction, and use SVM for two-category (win, lose) training and prediction.

[0188] Considering the switching and differences between the group stage (win, draw, lose) and the knockout stage (win, lose), we designed a two-layer classification model (multi-classification + two-classification) to predict the draw of the first-layer model, such as Figure 1 shown.

[0189] We tried several one-layer multi-classification models (win, draw, loss), and considering that many multi-classification models on the Tesla platform have not yet been deployed, we finally chose the XGboost model developed by Chen Tianqi, a scholar at the University of Washington. The basis is as follows:

[0190] XGboost is an efficient system implementation of Gradient Boosting, not a single algorithm. In addition to using tree (gbtree), the base learner in XGboost can also use linear classifier (gblinear). In this case, XGboost is equivalent to logistic regression (classification problem) or linear regression (regression problem) with L1 and L2 regularization terms.

[0191] XGboost performs a second-order Taylor expansion on the cost function and adds a regularization term to the cost function to control the complexity of the model. The regularization term includes the number of leaf nodes in the tree and the sum of the squares of the L2 norm of the score output on each leaf node, which reduces the variance of the model, making the learned model simpler and preventing overfitting.

[0192] XGboost automatically utilizes CPU multithreading, and appropriately improves gradient boosting, adds pruning, and controls the complexity of the model;

[0193] XGboost allows column (feature) subsampling to prevent overfitting, which draws on the idea of ​​Random Forest. This is also a feature of XGboost that is different from traditional GBDT.

[0194] An approximate algorithm for finding split nodes is implemented to speed up and reduce memory consumption. The node splitting algorithm can automatically exploit the sparsity of features.

[0195] We define the learning tasks and corresponding learning objectives in the model setting, and try to select the objective function as follows:

[0196] "multi:softmax" - Let XGBoost use the softmax objective function to handle multi-classification problems, and you need to set the parameter num_class (number of categories);

[0197] "multi:softprob" - Same as softmax, but the output is a vector of ndata*nclass, which can be converted into a matrix of ndata rows and nclass columns. Each row of data represents the probability of the sample belonging to each category.

[0198] The SVM used in the two-layer binary classification model makes binary predictions for the draws generated by the one-layer classification model in the group stage.

[0199] The core idea of ​​SVM is essentially similar to the linear regression LR classification method, which is to find a set of weight coefficients that can be classified after linear representation. We first use a set of training sets to train the weight coefficients in SVM, and then we can classify the test set.

[0200] Specifically, SVM first trains a separation hyperplane, and then the plane is the decision boundary of the classification, and the two categories are on both sides of the plane. Obviously, the classic SVM algorithm is only applicable to two-category classification problems. Of course, after improvement, SVM can also be applied to multi-class classification problems. We hope to find the points closest to the separation hyperplane and ensure that they are as far away from the separation plane as possible. The distance from the point to the separation plane is called the interval margin, and we hope that this margin is as large as possible. The support vector support vector is the points closest to the separation hyperplane, and we want to maximize the distance from the support vector to the separation plane. The specific solution process is as follows:

[0201] For the two-category classification problem, the training set T = {(x1, y1), (x2, y2), ⋯, (xN, yN)}, whose category yi∈{0, 1}, the linear SVM obtains the separation hyperplane by learning:

[0202]

[0203] And the corresponding classification decision function:

[0204]

[0205] Where sign is the sign function, mapping positive numbers to 1 and negative numbers to -1.

[0206] It is easy to get by calculation:

[0207]

[0208] The SVM classification problem can be described as maximizing (equivalent to minimizing 0.5∥w∥20.5∥w∥2); The constrained optimization problem of linear classification is transformed into the following convex optimization problem:

[0209]

[0210] In the linearly separable case, the instance of the sample point in the training data set that is closest to the separating hyperplane is called the support vector.

[0211] The support vectors satisfy:

[0212]

[0213] That is, the positive sample points satisfy:

[0214]

[0215] Negative sample points satisfy:

[0216]

[0217] Step 4: Confirm the classification results and provide prediction guidance. As shown in the following table:

[0218] Take the prediction of the 2016 European Cup quarter-finals as an example (label = 1 win, 2 losses, 0 draw. Since the quarter-finals must decide the winner, the results of the two-layer binary classification SVM model are used, and the one-layer three-classification XGboost model can be used for the group stage).

[0219]

[0220] Table 2 Example of output results of the two-layer binary classification SVM model

[0221] Compared with the predicted results, the four major upsets of the 2016 European Cup all came true (within 90 minutes), as shown in the following table (the predicted winning team and the winning probability of the winning team are given by the machine learning model): In the 1 / 8 finals, the big upset country P advanced; in the 1 / 8 finals, country L defeated country J and advanced;

[0222]

[0223] Table 3 Example 1 of machine learning model prediction results

[0224] In the quarter-finals, Country M defeated Country Q and advanced;

[0225]

[0226] Table 4 Example 2 of machine learning model prediction results

[0227] In the final, Country D tied with Country E;

[0228]

[0229] Table 5 Example 3 of machine learning model prediction results

[0230] Calculated based on the 64 games of the 2016 European Cup, the accuracy rate reached 78%, and all four upsets in the round of 16 were predicted.

[0231] The embodiment of the present invention also provides a game result prediction device based on machine learning, such as Fig.10 As shown, the device comprises:

[0232] The first determination module 1001 is used to determine the contestants of the target competition;

[0233] A second determination module 1003, configured to determine feature data of the contestant based on historical competition data of the contestant;

[0234] An acquisition module 1005 is used to acquire the game result interference data of the contestant;

[0235] The first prediction module 1007 is used to predict the game result of the feature data and the game result interference data based on the first game result prediction model to obtain the first game result data of the target game.

[0236] In some embodiments, the apparatus further comprises:

[0237] The second prediction module is used to predict the game result of the contestant's feature data and the game result interference data based on the second game result prediction model to obtain the second game result data of the target game.

[0238] In some embodiments, the acquisition module includes:

[0239] A first acquisition unit is used to acquire information about incidental events between contestants and contestants;

[0240] The first determining unit is used to determine the game result interference data based on the incident information.

[0241] In some embodiments, the first prediction module includes:

[0242] a first training participant acquisition unit, configured to acquire historical match data of a training participant with a match result of win, draw or loss, wherein the training participant includes the participant;

[0243] A first training participant characteristic data determining unit, configured to determine characteristic data of the training participant based on historical match data of win, draw and loss as the match results;

[0244] A first training contestant competition result interference data acquisition unit, used to acquire the competition result interference data of the training contestant;

[0245] The first match result prediction model training unit is used to train a preset machine learning model based on the feature data of the training contestants and the match result interference data to obtain a first match result prediction model.

[0246] In some embodiments, the apparatus further comprises:

[0247] The second prediction module is used to predict the game result of the contestant's feature data and the game result interference data based on the second game result prediction model to obtain the second game result data of the target game.

[0248] In some embodiments, the second prediction module includes:

[0249] a second training contestant acquisition unit, configured to acquire historical match data of a match result of a training contestant, the training contestant including the contestant;

[0250] A second training participant characteristic data determination unit, configured to determine the characteristic data of the training participant based on the historical match data of the match results of win or loss;

[0251] A second training contestant competition result interference data acquisition unit, used to acquire the competition result interference data of the training contestant;

[0252] The second game result prediction model training unit is used to train the preset machine learning model based on the feature data of the training contestants and the game result interference data to obtain the second game result prediction model.

[0253] In some embodiments, the acquisition module includes:

[0254] A first acquisition unit is used to acquire information about incidental events between contestants and contestants;

[0255] The first determining unit is used to determine the game result interference data based on the incident information.

[0256] In some embodiments, the second determining module includes:

[0257] A second determining unit, configured to determine the confrontation characteristic data of the team contestant based on the win-loss information in the historical match data of the team contestant, wherein the historical match data includes match data of matches within a preset time period;

[0258] A third determining unit, configured to determine the competition status data of the team competition object based on the historical competition data;

[0259] A second acquisition unit is used to acquire first characteristic data of a contestant in the team contestant object;

[0260] A third acquisition unit is used to acquire second characteristic data of the strategic instructor among the team contestants;

[0261] The first processing unit is used to use the confrontation feature data of the team contestant, the competition status data of the team contestant, the first feature data of the contestant and the second feature data of the strategic instructor as the feature data of the team contestant.

[0262] In some embodiments, the second determining module includes:

[0263] A fourth acquisition unit, configured to acquire score information of the game of the team contestants within the preset time period from the historical game data;

[0264] A first calculation unit is used to calculate the number of goals conceded and the entropy of goals conceded in the game of the team participating object within a preset time period according to the score information;

[0265] A fifth obtaining unit, configured to obtain, from the historical game data, ranking information of the team's competitors in the games within the preset time period;

[0266] A second calculation unit, configured to calculate ranking growth rate information of the team contestants according to the ranking information;

[0267] A sixth obtaining unit, configured to obtain the competition opponents of the team competition object from the historical competition data;

[0268] A fourth determining unit, configured to determine the strength information of the opponent;

[0269] The fifth determining unit is used to determine the competition status data of the team competition object according to the number of goals conceded, goal conceded entropy, ranking information and opponent strength information of the game within the preset time period.

[0270] In some embodiments, the second determining module includes:

[0271] A sixth obtaining unit, configured to obtain score information of a second number of recent games of the team contestants from the historical game data;

[0272] A third calculation unit is used to calculate the score standard deviation and score entropy of the team contestants in the second number of recent games according to the score information;

[0273] A seventh obtaining unit, configured to obtain, from the historical game data, information on the number of mistakes made by the team participants in a second number of recent games;

[0274] A fourth calculation unit, configured to calculate the error standard deviation and error entropy of the team contestant in the second most recent number of games according to the error number information;

[0275] A sixth determining unit, configured to determine, from the historical game data, game status data corresponding to a third number of recent games of the first preset game type;

[0276] A seventh determining unit, configured to determine the winning number information of the preset event type game according to the game status data corresponding to the preset event type game;

[0277] An eighth determination unit is used to determine the competition status data of the team participant based on the score standard deviation and score entropy, error standard deviation and error entropy of the team participant's most recent second number of games and the number of wins of games of a preset event type.

[0278] In some embodiments, the second determining module includes:

[0279] A ninth determining unit, configured to determine, based on the historical competition data, ranking information and points information of the participant in the event corresponding to the target competition;

[0280] A tenth determining unit, configured to determine game status data corresponding to a game of a second preset game type from the historical game data;

[0281] An eleventh determining unit is used to determine the competition status data of the team participant based on the ranking information, the points information and the competition status data.

[0282] In some embodiments, the second determining module includes:

[0283] An eighth obtaining unit, configured to obtain the game data of the team participants in the preliminary round from the game data of the preliminary round;

[0284] The second processing unit is used to use the game status data of the team contestant in the preliminary game as the competition status data of the team contestant.

[0285] In some embodiments, the second determining module includes:

[0286] a twelfth determining unit, configured to determine the confrontation characteristic data of the individual contestant based on the win-loss information in the historical match data of the individual contestant, wherein the historical match data includes match data of matches within a preset time period;

[0287] a thirteenth determining unit, configured to determine the competition status data of the individual competition object based on the historical competition data;

[0288] A ninth acquisition unit, configured to acquire third characteristic data of the individual contestant;

[0289] The third processing unit is used to use the confrontation feature data of the individual contestant, the competition status data of the individual contestant and the third feature data of the contestant as the feature data of the individual contestant.

[0290] In some embodiments, the acquisition module includes:

[0291] a tenth obtaining unit, configured to obtain information on incidental events between the contestants and between the contestants;

[0292] A fourteenth determining unit is used to determine the game result interference data based on the incidental event information.

[0293] The device and method embodiments in the device embodiments are based on the same inventive concept.

[0294] The present invention also provides a match result prediction device based on machine learning, the device comprising a processor and a memory, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the match result prediction method based on machine learning as described above.

[0295] In the embodiments of this specification, the memory can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, application programs required for functions, etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide the processor with access to the memory.

[0296] It can be seen from the embodiments of the method, device and apparatus for predicting game results based on machine learning provided by the present invention that the scheme of the embodiments of this specification uses the present invention to analyze the correlation between the game results and the team's strength, the status of the last N games, the coach, injuries and other factors:

[0297] 1) The factors that affect the results of the game are divided into strength and black swan. The strength part is further divided into four parts: "team", "coach", "player", and "historical confrontation". In addition to the pre-processing of basic data, each part also includes deep processing with statistical and physical significance (such as goal frequency, average number of goals / standard deviation in N games, etc.).

[0298] 2) Collect and extract historical battle data samples with sufficient dimensions, sufficient time depth, and sufficient noise.

[0299] At the same time, the method used in the present invention utilizes the powerful ability of machine learning, combines the dimensionless statistical features based on the output of statistical functions, and designs a reasonable multi-classification regression model that can output multiple classification labels. It can accurately quantify the probability of occurrence of the three results of win, loss and draw, and guide football betting.

[0300] It should be noted that the sequence of the embodiments of the present invention described above is for description only and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0301] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, system and server embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0302] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0303] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting game results based on machine learning, characterized in that: The method comprises: Determine the participants of the target competition; Determining characteristic data of the contestant based on historical competition data of the contestant; Obtaining game result interference data of the contestant; the game result interference data includes conflict information of the contestant, injury information of the contestant, media event information of the contestant, and ban information of the contestant; Predicting a match result on the feature data and the match result interference data based on a first match result prediction model to obtain first match result data of the target match; the first match result prediction model is a three-classification model, and the first match result data is one of a win, a loss, and a draw; In a case where the match result of the target match includes a win, a loss or a draw, the first match result data is used as the target match result data; the target match result data is used for predicting a first match scenario, and the first match scenario includes a group match scenario; When the match result of the target match includes a win or a loss, and the first match result data is a draw, the feature data and the match result interference data are used to predict the match result based on a second match result prediction model to obtain second match result data of the target match, and the second match result data is used as the target match result data; the second match result prediction model is a binary classification model, and the second match result data is one of a win or a loss; when the match result of the target match includes a win or a loss, the target match result data is used to predict a second match scenario, and the second match scenario includes a knockout scenario; Wherein, when the contestant is a team contestant, determining the feature data of the contestant based on the historical game data of the contestant includes: Determining the confrontation characteristic data of the team contestant based on the victory and defeat information in the historical match data of the team contestant, wherein the historical match data includes the match data of the matches within a preset time period; the confrontation characteristic data is determined based on the victory and defeat information of the matches played by the team contestant within the most recent preset number of years; Determine the competition status data of the team contestant based on the historical competition data; the competition status data includes fluctuation status data and fixed status data, the fluctuation status data is used to reflect the fluctuation of the team contestant in the historical competition process, and the fixed status data is used to reflect the static status of the team contestant in the corresponding competition at present; Acquire first characteristic data of a contestant among the team contestants; the first characteristic data is data that can reflect the strength of the contestant; Acquire the second characteristic data of the strategic instructor among the team participants; the second characteristic data is data that can reflect the strength of the strategic instructor; taking the confrontation characteristic data of the team contestant, the competition status data of the team contestant, the first characteristic data of the contestant and the second characteristic data of the strategic instructor as the characteristic data of the team contestant; When the contestant is an individual contestant, determining the characteristic data of the contestant based on the historical competition data of the contestant includes: Determining the confrontation characteristic data of the individual contestant based on the win-loss information in the historical match data of the individual contestant, wherein the historical match data includes the match data of the matches within a preset time period; Determining the competition status data of the individual competition object based on the historical competition data; Acquire the third characteristic data of the individual contestant; the third characteristic data is consistent with the first characteristic data of the contestant in the team contestant; The confrontation feature data of the individual contestant, the competition status data of the individual contestant and the third feature data of the contestant are used as feature data of the individual contestant.

2. The method according to claim 1, characterized in that: The first match result prediction model is established in the following way: Obtaining historical match data of a training contestant with a match result of win, draw or loss, wherein the training contestant includes the contestant; Determining characteristic data of the training contestant based on historical match data with match results of win, draw or loss; Obtaining game result interference data of the training contestant; The preset machine learning model is trained based on the characteristic data of the training contestants and the game result interference data to obtain a first game result prediction model.

3. The method according to claim 1, characterized in that: The second match result prediction model is established in the following way: Acquire historical competition data of a training contestant with a win or loss result, wherein the training contestant includes the contestant; Determining characteristic data of the training contestant based on historical competition data with win or loss results; Obtaining game result interference data of the training contestant; The preset machine learning model is trained based on the characteristic data of the training contestants and the interference data of the game results to obtain a second game result prediction model.

4. The method according to claim 1, characterized in that: The determining of the competition status data of the team competition object based on the historical competition data includes: Acquiring score information of the game of the team participants within the preset time period from the historical game data; Calculate the number of goals conceded and the goal conceded entropy of the team participants in the game within a preset time period according to the score information; Acquire ranking information of the team's competitors in the preset time period from the historical competition data; Calculate ranking growth rate information of the team participants according to the ranking information; Obtaining the opponents of the team's competitors from the historical game data; Determining strength information of the participating opponents; The competition status data of the team participants is determined according to the number of goals conceded, goal conceded entropy, ranking information and opponent strength information of the game within the preset time period.

5. The method according to claim 1, characterized in that The determining of the competition status data of the team competition object based on the historical competition data includes: Obtaining score information of a second number of recent games of the team participants from the historical game data; Calculate the score standard deviation and score entropy of the team's contestants in the second most recent number of games according to the score information; Acquire information on the number of mistakes made by the team participants in the second number of recent games from the historical game data; Calculate the error standard deviation and error entropy of the team's competition object in the second most recent number of games according to the error number information; Determining, from the historical game data, game status data corresponding to a third number of recent games of the first preset game type; Determine the winning number information of the preset event type game according to the game status data corresponding to the preset event type game; The competition status data of the team participant is determined based on the score standard deviation and score entropy, error standard deviation and error entropy of the team participant in the latest second number of games and the number of wins of the preset event type game.

6. The method according to claim 1, characterized in that The determining of the competition status data of the team competition object based on the historical competition data includes: Determine the ranking information and points information of the contestant in the event corresponding to the target game based on the historical game data; Determining game status data corresponding to a game of a second preset game type from the historical game data; The competition status data of the team participants is determined based on the ranking information, the points information and the competition status data.

7. The method according to claim 1, characterized in that When the historical match data includes match data of a preliminaries of the target match, determining the competition status data of the team competition object based on the historical match data includes: Acquire the game data of the team participants in the preliminary round from the game data of the preliminary round; The competition status data of the team contestants in the preliminary round is used as the competition status data of the team contestants.

8. A device for predicting game results based on machine learning, characterized in that: The device comprises: A first determination module is used to determine the participants of the target competition; A second determination module, configured to determine feature data of the contestant based on historical competition data of the contestant; An acquisition module, used to acquire the competition result interference data of the contestant; the competition result interference data includes the contradiction information of the contestant, the injury information of the contestant, the media event information of the contestant and the suspension information of the contestant; A first prediction module is used to predict the match result of the feature data and the match result interference data based on a first match result prediction model to obtain first match result data of the target match; the first match result prediction model is a three-classification model, and the first match result data is one of a win, a loss, and a draw; a third prediction module, configured to use the first match result data as target match result data when the match result of the target match includes win, loss or draw; the target match result data is used for prediction of a first match scenario, wherein the first match scenario includes a group match scenario; a second prediction module, configured to, when the match result of the target match includes a win or a loss and the first match result data is a draw, predict the match result of the feature data and the match result interference data based on a second match result prediction model to obtain second match result data of the target match, and use the second match result data as the target match result data; the second match result prediction model is a binary classification model, and the second match result data is one of a win or a loss; when the match result of the target match includes a win or a loss, the target match result data is used for prediction of a second match scenario, and the second match scenario includes a knockout scenario; The second determining module comprises: A second determining unit is used to determine the confrontation characteristic data of the team contestant based on the victory and defeat information in the historical match data of the team contestant, wherein the historical match data includes the match data of the match within a preset time period; the confrontation characteristic data is determined based on the victory and defeat information of the match played by the team contestant within the most recent preset number of years; A third determining unit is used to determine the competition status data of the team participating object based on the historical competition data; the competition status data includes fluctuation status data and fixed status data, the fluctuation status data is used to reflect the fluctuation of the team participating object in the historical competition process, and the fixed status data is used to reflect the static status of the team participating object in the corresponding competition at present; The second acquisition unit is used to acquire first characteristic data of a contestant in the team contestant; the first characteristic data is data that can reflect the strength of the contestant; The third acquisition unit is used to acquire the second characteristic data of the strategic instructor among the team participants; the second characteristic data is data that can reflect the strength of the strategic instructor himself; A first processing unit is used to use the confrontation feature data of the team contestant, the competition status data of the team contestant, the first feature data of the contestant and the second feature data of the strategic instructor as feature data of the team contestant; The second determining module comprises: a twelfth determination unit, configured to determine the confrontation characteristic data of the individual contestant based on the win-loss information in the historical match data of the individual contestant, wherein the historical match data includes match data of matches within a preset time period; a thirteenth determining unit, configured to determine the competition status data of the individual competition object based on the historical competition data; A ninth acquisition unit, configured to acquire third characteristic data of the individual contestant; the third characteristic data is consistent with the first characteristic data of the contestant in the team contestant; The third processing unit is used to use the confrontation feature data of the individual contestant, the competition status data of the individual contestant and the third feature data of the contestant as the feature data of the individual contestant.

9. The device according to claim 8, characterized in that The first prediction module comprises: a first training participant acquisition unit, configured to acquire historical match data of a training participant with a match result of win, draw or loss, wherein the training participant includes the participant; A first training participant characteristic data determining unit, configured to determine characteristic data of the training participant based on historical match data of win, draw and loss as the match results; A first training contestant competition result interference data acquisition unit, used to acquire the competition result interference data of the training contestant; The first match result prediction model training unit is used to train a preset machine learning model based on the feature data of the training contestants and the match result interference data to obtain a first match result prediction model.

10. The device according to claim 8, characterized in that The second prediction module comprises: a second training contestant acquisition unit, configured to acquire historical match data of a match result of a training contestant, the training contestant including the contestant; A second training participant characteristic data determination unit, configured to determine the characteristic data of the training participant based on the historical match data of the match results of win or loss; A second training contestant competition result interference data acquisition unit, used to acquire the competition result interference data of the training contestant; The second game result prediction model training unit is used to train the preset machine learning model based on the feature data of the training contestants and the game result interference data to obtain the second game result prediction model.

11. The device according to claim 8, characterized in that The third determining unit includes: A fourth acquisition unit, configured to acquire score information of the game of the team contestants within the preset time period from the historical game data; A first calculation unit is used to calculate the number of goals conceded and the entropy of goals conceded in the game of the team participating object within a preset time period according to the score information; A fifth obtaining unit, configured to obtain, from the historical game data, ranking information of the team's competitors in the games within the preset time period; A second calculation unit, configured to calculate ranking growth rate information of the team contestants according to the ranking information; A sixth obtaining unit, configured to obtain the competition opponents of the team competition object from the historical competition data; A fourth determining unit, configured to determine the strength information of the opponent; The fifth determining unit is used to determine the competition status data of the team competition object according to the number of goals conceded, goal conceded entropy, ranking information and opponent strength information of the game within the preset time period.

12. The device according to claim 8, characterized in that The third determining unit includes: A sixth obtaining unit, configured to obtain score information of a second number of recent games of the team contestants from the historical game data; A third calculation unit is used to calculate the score standard deviation and score entropy of the team contestants in the second number of recent games according to the score information; A seventh obtaining unit, configured to obtain, from the historical game data, information on the number of mistakes made by the team participants in a second number of recent games; A fourth calculation unit, configured to calculate the error standard deviation and error entropy of the team contestant in the second most recent number of games according to the error number information; A sixth determining unit, configured to determine, from the historical game data, game status data corresponding to a third number of recent games of the first preset game type; A seventh determining unit, configured to determine the winning number information of the preset event type game according to the game status data corresponding to the preset event type game; An eighth determination unit is used to determine the competition status data of the team participant based on the score standard deviation and score entropy, error standard deviation and error entropy of the team participant's most recent second number of games and the number of wins of games of a preset event type.

13. The device according to claim 8, characterized in that The third determining unit includes: A ninth determining unit, configured to determine, based on the historical competition data, ranking information and points information of the participant in the event corresponding to the target competition; A tenth determining unit, configured to determine game status data corresponding to a game of a second preset game type from the historical game data; An eleventh determining unit is used to determine the competition status data of the team participant based on the ranking information, the points information and the competition status data.

14. The device according to claim 8, characterized in that The third determining unit includes: An eighth obtaining unit, configured to obtain the game data of the team participants in the preliminary round from the game data of the preliminary round; The second processing unit is used to use the game status data of the team contestant in the preliminary game as the competition status data of the team contestant.

15. A game result prediction device based on machine learning, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the method for predicting game results based on machine learning as described in any one of claims 1 to 7.

Images