Prediction method, device, equipment and readable storage medium based on feature collision

Abstract

The present invention provides a prediction method, device, equipment and readable storage medium based on feature collision, including: obtaining multiple samples, performing feature extraction on them, and forming a feature set according to the analysis target; determining the number k of features extracted from the feature set each time, extracting k features, and forming a first feature branch based on the k features; putting the selected k features back into the feature set, and then reselecting k features from the feature set, judging whether the reselected k features are the same as the previously selected features, and if so, accumulating the number of first feature branches, and if not, establishing a second feature branch based on the reselected k features, after repeating a preset number of times, forming a branch group, and determining multiple key features in the branch group; using the key features to perform feature anchor matching on user information, and generating a user prediction result based on the matching data. This improves the problem of large single-dimensional error volatility and low noise resistance in existing prediction methods.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a feature collision-based prediction method, device, equipment and readable storage medium. Background Art

[0002] In some existing medical systems, information in historical patient medical records is processed. For example, feature extraction is performed on historical patient medical records. The collected patient information can be analyzed and compared with the features of historical patient medical records to predict the patient's physical condition.

[0003] However, in the existing technology, after feature extraction, there is no intersection between the branches of each feature, which makes the branch results easily one-sided, that is, falling into a single linear law. The single law is greatly affected by data volatility and can easily lead to misjudgment, making the prediction results easily misleading patients or medical staff.

[0004] In view of this, this application is filed. Summary of the Invention

[0005] The present invention discloses a prediction method, device, equipment and readable storage medium based on feature collision, aiming to improve the problems of large single-dimensional error volatility and low noise resistance in existing prediction methods.

[0006] The first embodiment of the present invention provides a method for predicting collision based on features, comprising:

[0007] Acquire multiple samples, perform feature extraction on the multiple samples, and form a feature set according to an analysis target;

[0008] Determining the number k of features extracted from the feature set each time, selecting k features from the feature set, and forming a first feature branch based on the k features;

[0009] Putting the k selected features back into the feature set, reselecting k features from the feature set, determining whether the reselected k features are the same as the previously selected features, and if so, accumulating the number of first feature branches; if not, establishing a second feature branch based on the reselected k features, and after repeating this for a preset number of times, forming a branch group, and determining a plurality of key features in the branch group;

[0010] Obtaining user information of the user to be predicted, performing feature anchor matching on the user information according to the plurality of key features, and generating matching data;

[0011] A prediction result of the user is generated based on the matching data.

[0012] Preferably, the step of extracting k features from the feature set each time is as follows:

[0013] Acquire non-parametric features that are not extracted from each sample multiple times, and the classification results corresponding to each non-parametric feature;

[0014] Determine the classification results of the samples by voting;

[0015] According to the classification result, it is determined that when the error branch rate of the non-parameter feature is minimized, the number of features extracted from the feature set is defined as k.

[0016] Preferably, the determining of the multiple key features in the branch group is specifically:

[0017] Calculate the initial wrong branch rate oui(1) of the characteristic branches in the branch group;

[0018] Select a feature from the branch group, randomly add noise interference to the feature, and calculate the error branch rate oui(2) of the feature branch again;

[0019] Calculate the importance of the feature, wherein the feature importance model is: feature importance IMP = ∑(oui(2)-oui(1)) / (number of branches in the branch group)

[0020] Repeat the above steps until the importance of each feature in the branch group is calculated;

[0021] The importance of obtaining features IMP is greater than the preset threshold value of multiple key features.

[0022] Preferably, generating a user prediction result based on the matching data is specifically:

[0023] Averaging the probabilities of each category of the matching data to generate an average value corresponding to each category;

[0024] The user's prediction result is generated according to the average value corresponding to each category.

[0025] A second embodiment of the present invention provides a feature-based collision prediction device, comprising:

[0026] a feature extraction unit, configured to obtain a plurality of samples, extract features from the plurality of samples, and form a feature set according to an analysis target;

[0027] a feature quantity determination unit, configured to determine the number k of features to be extracted from the feature set each time, select k features from the feature set, and form a first feature branch based on the k features;

[0028] a key feature determination unit, configured to place the k selected features back into the feature set, reselect k features from the feature set, determine whether the k reselected features are the same as the previously selected features, and if so, accumulate the number of first feature branches; if not, establish a second feature branch based on the k reselected features, form a branch group after repeating this for a preset number of times, and determine a plurality of key features in the branch group;

[0029] A user information acquisition unit, configured to acquire user information of a user to be predicted, perform feature anchor matching on the user information according to a plurality of key features, and generate matching data;

[0030] The prediction result generating unit is used to generate the user's prediction result according to the matching data.

[0031] Preferably, the feature quantity determination unit is specifically configured to:

[0032] Acquire the non-parametric features that are not extracted from each sample multiple times, and the classification status corresponding to each non-parametric feature;

[0033] Determine the classification results of the samples by voting;

[0034] According to the classification result, it is determined that when the error branch rate of the non-parameter feature is minimized, the number of features extracted from the feature set is defined as k.

[0035] Preferably, the key feature determination unit is specifically configured to:

[0036] Calculate the initial error branch rate oui(1) of the characteristic branches in the branch group;

[0037] Select a feature from the branch group, randomly add noise interference to the feature, and calculate the error branch rate oui(2) of the feature branch again;

[0038] Calculate the importance of the feature, wherein the feature importance model is: feature importance IMP = ∑(oui(2)-oui(1)) / (number of branches in the branch group)

[0039] Repeat the above steps until the importance of each feature in the branch group is calculated;

[0040] The importance of obtaining features IMP is greater than the preset threshold value of multiple key features.

[0041] Preferably, the prediction result generating unit is specifically used to:

[0042] Averaging the probabilities of each category of the matching data to generate an average value corresponding to each category;

[0043] The user's prediction result is generated according to the average value corresponding to each category.

[0044] The third embodiment of the present invention provides a feature-collision-based prediction device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a feature-collision-based prediction method as described in any one of the above items.

[0045] A fourth embodiment of the present invention provides a readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer readable storage medium is located to implement a feature-based collision prediction method as described in any one of the above.

[0046] Based on the feature collision-based prediction method, device, equipment and readable storage medium provided by the present invention, features are first extracted from multiple samples, and a feature set is formed according to the analysis target; then, the number k of features extracted from the feature set each time is determined by estimating the error branch rate to ensure that the results of those non-participating samples do not deviate much when participating and not participating, and a branch group is formed by randomly colliding and then putting the feature set back and colliding again, and multiple key features in the branch group are determined. Finally, user information of the user to be predicted is obtained, and feature anchor matching is performed on the user information according to the multiple key features, and matching data is generated; the user's prediction result is generated according to the matching data, thereby improving the problems of large single-dimensional error volatility and low noise resistance in the existing prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flowchart of a feature-based collision prediction method provided by the first embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of the process of determining the number of features extracted from a feature set each time provided by the present invention.

[0049] Figure 3 1 is a schematic diagram of a module of a feature-based collision prediction device provided by the first embodiment of the present invention; DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 efforts are within the scope of protection of the present invention.

[0051] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0052] It should be understood that the embodiments described are only a portion 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 persons of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0053] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0054] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0055] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0056] The "first" and "second" mentioned in the embodiments are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It is understood that the specific order or precedence of "first" and "second" can be interchanged where appropriate. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein.

[0057] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0058] The present invention discloses a prediction method, device, equipment and readable storage medium based on feature collision, aiming to improve the problems of large single-dimensional error volatility and low noise resistance in existing prediction methods.

[0059] See also Figure 1The first embodiment of the present invention provides a feature collision-based prediction method, which can be executed by a feature collision prediction device (hereinafter referred to as the prediction device), and in particular, by one or more processors in the prediction device to implement the following steps:

[0060] S101, acquiring multiple samples, performing feature extraction on the multiple samples, and forming a feature set according to an analysis target;

[0061] In this embodiment, the prediction device may be a device located at a terminal (such as a terminal computer, smart phone, tablet computer, etc. with data processing capabilities), which can communicate data with the medical system, wherein the medical system may be a cloud database or a terminal database, which stores historical medical data.

[0062] In particular, in this embodiment, the sample may be information of multiple data of patients suffering from a certain disease in a medical system. The information may include the patient's name, age, gender, heart rate status, and corresponding disease status. When extracting features, age, gender, and heart rate status may be regarded as features, and a feature set may be formed with the disease status as the analysis target.

[0063] S102, determining the number k of features to be extracted from the feature set each time, selecting k features from the feature set, and forming a first feature branch based on the k features;

[0064] It's important to note that when establishing feature branches, the greater the correlation between any two branches in the branch group, the greater the prediction error rate. However, if the number of features in a branch increases, the branches become more dispersed, and the overall error rate also increases. The above problems often occur when the traditional approach of subjectively selecting the k value.

[0065] In this embodiment, a more appropriate number of k is determined by calculating the minimum collision error rate. It should be understood that the minimum collision error rate includes features that do not participate in sample extraction into the calculation to see the probability of their impact on the final result prediction deviation. The smaller the probability, the more it indicates that the number of features currently taken from each sample is sufficient to determine the entire result, that is, samples that do not participate in the prediction will have little impact on the final result even if they participate in the prediction, so k is selected appropriately at this time.

[0066] Specifically: See Figure 2 ;

[0067] S1021, obtaining non-parametric features that are not extracted from each sample multiple times, and the classification status corresponding to each non-parametric feature;

[0068] It should be understood that in the process of feature extraction, about 1 / 3 of the features will not be extracted during the establishment of each feature branch. These features are the non-participating extraction features of each feature branch (i.e., non-parametric features). The specific calculation process is: We temporarily call this 1 / 3 of the data non-parametric features OUI. We can estimate the OUI error branch rate;

[0069] For each sample, calculate its classification based on the branch of the non-parametric feature OUI sample, and calculate the branch of the feature branch based on its corresponding OUI feature. (If the sampled samples are systematically labeled, then the unlabeled samples, i.e., the unsampled samples, are classified based on the feature distribution and the branches formed by the unsampled samples. For example, if sample 1 has three features a=1, b=2, and c=3, the branch for sample 1 is high blood sugar, so a=1, b=2, and c=3 are the characteristics of high blood sugar.)

[0070] S1022, determining the classification result of the sample by voting;

[0071] S1023 , determining, based on the classification result, the number of features extracted from the feature set when the error branching rate of the non-parameter feature is minimized, and defining it as k.

[0072] It should be noted that the oui error branching rate is calculated as the ratio of the number of feature samples with incorrect classification to the total number of feature samples. A classification error refers to the following: after sampling n times in steps S1021 and S1022, if 50 samples are extracted and the majority rule is applied, 48 samples may be classified as classification result a, and the other 2 as classification result b. Therefore, the oui error branching rate is 2 / 50. A correct classification is indicated, and b is an incorrect classification. A smaller error branching rate indicates that the deviation between the results of the samples that did not participate and those that did not participate is not large, so k is more appropriate for prediction.

[0073] The oui false branch rate is an unbiased estimator that can be evaluated internally, meaning that an unbiased estimate of the error can be established during the branch generation process. (The mathematical expectation of an estimator is equal to the true value of the estimated parameter, which is unbiased and is a criterion for evaluating the quality of an estimator.)

[0074] S103, putting the k selected features back into the feature set, and then re-selecting k features from the feature set, and judging whether the k re-selected features are the same as the previously selected features, if so, accumulating the number of first feature branches, and if not, establishing a second feature branch based on the k re-selected features, and after repeating the preset number of times, forming a branch group, and determining multiple key features in the branch group.

[0075] It should be noted that, in this embodiment, a branch group formed by n feature branches is formed by random collision, replacement and then collision. The reason why it is called random collision is that it is similar to a small ball falling into a ball box, which will knock itself out or knock others out. This process is random and uncertain, so it is called random collision. It should be understood that random collision: random collision can make the extraction results diverse. If it is not a random extraction, but a regular extraction, it will lead to the same feature sample values. Replacement and collision: If the traditional no-replacement extraction is used, each feature branch will have no intersection, which will make the branch results easily one-sided and lack fairness. This design collects, replaces and then collisions. Fairness can be guaranteed, and the results can be predicted by the cumulative number later, and over-data fitting can be prevented.

[0076] S104, obtaining user information of the user to be predicted, performing feature anchor matching on the user information according to the plurality of key features, and generating matching data;

[0077] In this embodiment, the initial erroneous branch rate oui(1) of the characteristic branches in the branch group is calculated;

[0078] Select a feature from the branch group, randomly add noise interference to the feature, and calculate the error branch rate oui(2) of the feature branch again;

[0079] Calculate the importance of the feature, wherein the importance model of the feature is: feature importance IMP = ∑(oui(2)-oui(1)) / (number of branches in the branch group);

[0080] Repeat the above steps until the importance of each feature in the branch group is calculated;

[0081] The importance of obtaining features IMP is greater than the preset threshold value of multiple key features.

[0082] It should be noted that the reason for designing the importance of the measurement feature in this way is that when noise is randomly added to a feature, if the oui increases, it means that this feature has a greater impact on the result of sample classification, indicating that it is more important.

[0083] S105: Generate a prediction result of the user based on the matching data.

[0084] In this embodiment, the probability of each category of the matching data is averaged to generate an average value corresponding to each category;

[0085] The user's prediction result is generated according to the average value corresponding to each category.

[0086] Based on the above embodiment, a prediction method based on feature collision is provided to improve the problems of large single-dimensional error volatility and low noise resistance in existing prediction methods.

[0087] Here is a simple example to help you understand this technical solution:

[0088] 1. Iteratively extract feature information based on 15 patient sample archival data.

[0089] During the statistical sampling process, sample feature extraction uses a random collision method, which involves random selection followed by replacement, followed by random selection again to ensure data fairness. Specifically, for example, if there are 15 patients, sampling is repeated 25 times, with each operation using a random collision method (randomly removing and then replacing) to obtain the total sample set. (For example, if k = 1, one feature is sampled each time).

[0090] For example, in sample 1, patient Zhang San has type 1 diabetes. His characteristics are age 27 and gender male. At this time, a feature of Zhang San, "age 27", is extracted randomly.

[0091]

[0092]

[0093] The reason why proportion is used in age in Table 1 is to convert the number of votes into the proportion of the cumulative number of votes for this feature in the total number of votes for this feature.

[0094] Sample 2: Li Si is a patient with type 2 diabetes. His characteristics are: male gender, history of the disease [yes], and heart rate grade 2. At this time, one of Li Si's characteristics "heart rate grade 2" is detected.

[0095] Each time a sample is taken, k features are taken, and the value of k is determined by the k value corresponding to the minimum collision error rate. The following feature judgments are obtained (the judgment is the type of diabetes that the entire patient has corresponding to the feature).

[0096] 2. After repeated extraction in step 1, a total of 5 feature classifications are obtained (i.e., feature set m). They are age, gender, degree of blood sugar abnormality, history of illness, and heart rate classification. The probability of each feature in the corresponding judgment result is calculated. This means that the first five tables in Explanation 5 below, [Table 1] to [Table 5], are:

[0097] 3. Before sampling, it is necessary to calculate the appropriate value of k. (Reducing the number of features selected, k, will reduce the relevance and classification ability of the branch tree; increasing k will increase both. Therefore, the key question is how to choose the optimal k. The optimal value is determined when k = 1 by using the minimum collision error rate method.)

[0098] We will calculate the value of k that minimizes the collision error rate by calculating the sample data. Since there are five features in the sample data, the program calculates k from k = 1 to k = 5, calculating the collision error rate for each k and then taking the k value that minimizes the collision error rate. Since k = 1 has the lowest collision error rate, we determine that k = 1 is the optimal value.

[0099] Since there is only one feature, it is the important feature. If there are multiple features, the important feature is determined by calculating the OUI under noise interference:

[0100] 4. Calculate the feature probabilities under different conditional branches as follows:

[0101] Table 1 Feature Branch [Age]:

[0102]

[0103] Table 2 Feature Branch [Gender]:

[0104]

[0105] Table 3 Feature branch [Abnormal blood sugar level]:

[0106]

[0107] Table 4 Feature branch [whether there is a medical history]:

[0108]

[0109] Table 5 Feature branches [heart rate classification]:

[0110]

[0111] The diabetes type of a patient predicted without diagnosis is as follows:

[0112] Age: 33 | Gender: Female | Severe blood sugar abnormality: Very severe | History of disease: Yes | Heart rate classification: Level 2

[0113] According to the anchor matching results, the patient's data matching is obtained, and then the prediction results of the patient are regressed to the mean value as follows:

[0114]

[0115] Finally, based on the results, it is predicted that the patient may be a patient with diabetes type 1.

[0116] See also Figure 3The second embodiment of the present invention provides a prediction device based on feature collision, comprising:

[0117] a feature extraction unit, configured to obtain a plurality of samples, perform feature extraction on the plurality of samples, and form a feature set according to an analysis target;

[0118] a feature quantity determination unit, configured to determine the number k of features to be extracted from the feature set each time, select k features from the feature set, and form a first feature branch based on the k features;

[0119] a key feature determination unit, configured to place the k selected features back into the feature set, reselect k features from the feature set, determine whether the k reselected features are the same as the previously selected features, and if so, accumulate the number of first feature branches; if not, establish a second feature branch based on the k reselected features, form a branch group after repeating this for a preset number of times, and determine a plurality of key features in the branch group;

[0120] A user information acquisition unit, configured to acquire user information of a user to be predicted, perform feature anchor matching on the user information according to a plurality of key features, and generate matching data;

[0121] The prediction result generating unit is used to generate the user's prediction result according to the matching data.

[0122] Preferably, the feature quantity determination unit is specifically configured to:

[0123] Acquire the non-parametric features that are not extracted from each sample multiple times, and the classification status corresponding to each non-parametric feature;

[0124] Determine the classification results of the samples by voting;

[0125] According to the classification result, it is determined that when the error branch rate of the non-parameter feature is minimized, the number of features extracted from the feature set is defined as k.

[0126] Preferably, the key feature determination unit is specifically configured to:

[0127] Calculate the initial error branch rate oui(1) of the characteristic branches in the branch group;

[0128] Select a feature from the branch group, randomly add noise interference to the feature, and calculate the error branch rate oui(2) of the feature branch again;

[0129] Calculate the importance of the feature, wherein the feature importance model is: feature importance IMP = ∑(oui(2)-oui(1)) / (number of branches in the branch group)

[0130] Repeat the above steps until the importance of each feature in the branch group is calculated;

[0131] The importance of obtaining features IMP is greater than the preset threshold value of multiple key features.

[0132] Preferably, the prediction result generating unit is specifically used for:

[0133] Averaging the probabilities of each category of the matching data to generate an average value corresponding to each category;

[0134] The user's prediction result is generated according to the average value corresponding to each category.

[0135] The third embodiment of the present invention provides a feature-collision-based prediction device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a feature-collision-based prediction method as described in any one of the above items.

[0136] A fourth embodiment of the present invention provides a readable storage medium storing a computer program, wherein the computer program can be executed by a processor of a device where the computer readable storage medium is located to implement a feature-based collision prediction method as described in any one of the above.

[0137] Based on the feature collision-based prediction method, device, equipment and readable storage medium provided by the present invention, features are first extracted from multiple samples, and a feature set is formed according to the analysis target; then, the number k of features extracted from the feature set each time is determined by estimating the error branch rate to ensure that the results of those non-participating samples do not deviate much when participating and not participating, and a branch group is formed by randomly colliding and then putting the feature set back and colliding again, and multiple key features in the branch group are determined. Finally, user information of the user to be predicted is obtained, and feature anchor matching is performed on the user information according to the multiple key features, and matching data is generated; the user's prediction result is generated according to the matching data, thereby improving the problems of large single-dimensional error volatility and low noise resistance in the existing prediction method.

[0138] For example, the computer programs described in the third and fourth embodiments of the present invention may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the device for implementing a feature-based collision prediction. For example, the device described in the second embodiment of the present invention.

[0139] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the feature-based collision prediction method, and utilizes various interfaces and lines to connect the various parts of the feature-based collision prediction method.

[0140] The memory can be used to store the computer program and / or module, and the processor implements various functions of a feature collision-based prediction method by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, a text conversion function, etc.); the data storage area can store data created based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0141] Wherein, if the implemented module is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of each of the above-mentioned method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0142] It should be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive effort.

[0143] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A prediction method based on feature collision, characterized in that: include: Acquire multiple samples, perform feature extraction on the multiple samples, and form a feature set according to an analysis target; wherein the samples are patient data information in a medical system; Determining the number k of features to be extracted from the feature set each time, selecting k features from the feature set, and forming a first feature branch based on the k features; the determining the number of features to be extracted from the feature set each time is specifically as follows: Acquire non-parametric features that are not extracted from each sample multiple times, and the classification results corresponding to each non-parametric feature; Determine the classification results of the samples by voting; According to the classification result, determining that the number of features extracted from the feature set is k when the error branch rate of the non-parameter feature is minimized; Putting the k selected features back into the feature set, reselecting k features from the feature set, determining whether the reselected k features are the same as the previously selected features, and if so, accumulating the number of first feature branches; if not, establishing a second feature branch based on the reselected k features, and after repeating this for a preset number of times, forming a branch group, and determining a plurality of key features in the branch group; Obtaining user information of the user to be predicted, performing feature anchor matching on the user information according to the plurality of key features, and generating matching data; A prediction result of the user is generated based on the matching data.

2. A feature-based collision prediction method according to claim 1, characterized in that: The determining of the multiple key features in the branch group is specifically: Calculate the initial wrong branch rate oui (1) of the characteristic branches in the branch group; Select a feature from the branch group, randomly add noise interference to the feature, and calculate the error branch rate oui (2) of the feature branch again; Calculate the importance of the feature, where the feature importance model is: feature importance IMP = ∑(oui(2)-oui(1)) / (number of branches in the branch group) Repeat the above steps until the importance of each feature in the branch group is calculated; The importance of obtaining features IMP is greater than the preset threshold value of multiple key features.

3. The feature-based collision prediction method according to claim 1, characterized in that: The generating of the user's prediction result based on the matching data is specifically as follows: Averaging the probabilities of each category of the matching data to generate an average value corresponding to each category; The user's prediction result is generated according to the average value corresponding to each category.

4. A prediction device based on feature collision, characterized in that: include: A feature extraction unit, configured to obtain a plurality of samples, extract features from the plurality of samples, and form a feature set according to an analysis target; wherein the samples are patient data information in a medical system; A feature quantity determination unit is configured to determine the number k of features to be extracted from the feature set each time, select k features from the feature set, and form a first feature branch based on the k features; the determination of extracting k features from the feature set each time is specifically as follows: Acquire the non-parametric features that are not extracted from each sample multiple times, and the classification status corresponding to each non-parametric feature; Determine the classification results of the samples by voting; According to the classification result, determining that the number of features extracted from the feature set is k when the error branch rate of the non-parameter feature is minimized; a key feature determination unit, configured to place the k selected features back into the feature set, reselect k features from the feature set, determine whether the k reselected features are the same as the previously selected features, and if so, accumulate the number of first feature branches; if not, establish a second feature branch based on the k reselected features, form a branch group after repeating this for a preset number of times, and determine a plurality of key features in the branch group; A user information acquisition unit, configured to acquire user information of a user to be predicted, perform feature anchor matching on the user information according to a plurality of key features, and generate matching data; The prediction result generating unit is used to generate the user's prediction result according to the matching data.

5. The feature-based collision prediction device according to claim 4, characterized in that: The key feature determination unit is specifically configured to: Calculate the initial wrong branch rate oui (1) of the characteristic branches in the branch group; Select a feature from the branch group, randomly add noise interference to the feature, and calculate the error branch rate oui (2) of the feature branch again; Calculate the importance of the feature, where the feature importance model is: feature importance IMP = ∑(oui(2)-oui(1)) / (number of branches in the branch group) Repeat the above steps until the importance of each feature in the branch group is calculated; The importance of obtaining features IMP is greater than the preset threshold value of multiple key features.

6. The feature-based collision prediction device according to claim 4, characterized in that: The prediction result generating unit is specifically used for: Averaging the probabilities of each category of the matching data to generate an average value corresponding to each category; The user's prediction result is generated according to the average value corresponding to each category.

7. A prediction device based on feature collision, characterized in that: The system comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a feature-based collision prediction method according to any one of claims 1 to 3.

8. A readable storage medium, characterized in that: A computer program is stored, and the computer program can be executed by a processor of the device where the readable storage medium is located to implement a feature-based collision prediction method as described in any one of claims 1 to 3.

Images