Timing sequence enhanced tree model classification algorithm
By introducing a timing enhancement algorithm in the tree model, using Gini exponent and timing forgetting function to fuse timing information, the problem of traditional tree models failing to effectively process timing data is solved, and prediction accuracy and data utilization are improved.
Patent Information
- Application Number
- CN202411478547.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-02-14
AI Technical Summary
Traditional tree models fail to effectively integrate timing information when processing timing data, resulting in logical function errors, affecting design performance, and increasing the difficulty of troubleshooting.
A time-series enhanced tree model classification algorithm is proposed. By obtaining training data and verification data, the tree model algorithm is used to train the initial tree model, and combined with the Gini index, maximum memory length and timing forget function, the adaptive compensation of timing information is realized, and finally obtained The final classification result after the timing enhancement.
This algorithm can effectively integrate timing information in the prediction process of classification results, improve the accuracy of prediction and the utilization of data information, and solve the problem that traditional tree models fail to effectively process timing data.
Smart Images
Figure CN119442050A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of machine learning algorithms, and in particular to a temporal enhanced tree model classification algorithm. Background Art
[0002] The tree model is a type of algorithm model based on a binary tree or multi-branch tree structure, and its basic structure consists of nodes and directed edges. The basic principle of the tree model is to first screen the feature items, traverse the sample data in the node to split, select the best split point, gradually split, improve the classification purity of the child nodes, and finally achieve sample classification.
[0003] A single tree model can be integrated to construct multiple trees to form an integrated learning model. Common tree model integration methods include Boosting and Bagging. Boosting is to first train a weak learner based on the sample, and then gradually improve the performance of the weak learner on the sample to a strong learner. Representative algorithms include XGBoost, GBDT, etc. Bagging adopts the parallel idea, first trains multiple learners, and then outputs the final result based on the results of multiple learners by voting or averaging. Representative algorithms include random forest, etc.
[0004] The tree model has the characteristics of strong interpretability and high computational efficiency, and is widely used in various engineering environments. However, for time series data, the traditional tree model predicts through the characteristic value of each sample point, without considering the integration of time series information. If the time series information is not integrated, it is easy to cause logical function errors in the application, affect the overall performance of the design, increase the difficulty of fault diagnosis, affect performance optimization decisions, affect time-dependent applications and services, affect the accuracy and completeness of data recovery, reduce query performance, increase the consumption of computing resources, and other problems. Summary of the invention
[0005] Based on this, it is necessary to provide a time-series enhanced tree model classification algorithm that can realize adaptive compensation of tree model time-series information, etc., in order to address the problem that traditional tree models do not integrate time-series information.
[0006] A temporal enhanced tree model classification algorithm, comprising the steps of:
[0007] Obtain training data and validation data;
[0008] Using a tree model algorithm to train the training data to obtain an initial tree model;
[0009] Input the verification data into the initial tree model, output the tree model prediction result, and obtain the Gini index of the output node corresponding to the tree model prediction result;
[0010] Calculate the weight coefficient of the tree model prediction result according to the Gini index;
[0011] Set the maximum memory gate length to limit the maximum memory length of timing information;
[0012] The actual memory length is calculated according to the maximum memory length, the Gini index and the weight coefficient of the tree model prediction result;
[0013] Setting a temporal forgetting function according to the actual memory length;
[0014] Using the temporal forgetting function to perform temporal forgetting on the samples in the memory gate to obtain a temporal prediction result;
[0015] The final classification result after time series enhancement is calculated based on the tree model prediction result and the time series prediction result.
[0016] In one embodiment, the sample set of verification data is D={(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x N ,y N )};in, is the feature vector, n is the number of features, y i ∈{1,2,…,K} is the classification mark, K is the total number of classifications, i=1,2,…,N, N is the sample size, and the sample point (x i ,y i ) is o i ;
[0017] The tree model prediction result corresponding to the sample set of the verification data is V Tree = {r (Tree,1) ,r (Tree,2) ,…r (Tree,N)}, each sample point o i The corresponding tree model prediction result is
[0018] in, is the sample point o i The corresponding probability distribution result of the kth classification.
[0019] In one embodiment, when the initial tree model is a single tree model or a bagging ensemble tree model, the following formula is used to calculate:
[0020]
[0021] In the above formula, is the predicted sample point o i The number of samples corresponding to the kth class of the corresponding leaf node, k = 1, 2, ... K; or
[0022] When the initial tree model is a Boosting ensemble tree model, each sample point o corresponding to the Boosting ensemble tree model with K classifications and C trees i The probability distribution of Let r (Tree,i) =P i ;
[0023] Calculate the sample point o according to the following Softmax function formula i The probability distribution results of
[0024]
[0025] Calculate the original score logit corresponding to category k according to the following formula k :
[0026]
[0027] Among them, T k,c (o i ) is the sample point o i The output data corresponding to the c-th tree model of category k.
[0028] In one embodiment, the step of obtaining the Gini index of the output node corresponding to the tree model prediction result is: calculating the Gini index Gini(o i ):
[0029]
[0030] In one embodiment, the step of calculating the weight coefficient of the initial tree model result according to the Gini index is: according to the following formula:
[0031] If and only if When Gini (o i ) to obtain the maximum value To determine the weight coefficient of the tree model prediction results
[0032] In one embodiment, the step of calculating the actual memory length according to the maximum memory length and the Gini index comprises the steps of:
[0033] Set the sample point o i The memory length L i As shown in the following formula:
[0034]
[0035] The memory length L is calculated using the following formula: i Correction is made to obtain the actual memory length l i :
[0036]
[0037] Among them, l M The maximum memory length is set.
[0038] In one embodiment, the temporal forgetting function f (t,i) Satisfy the conditions shown in the following formula:
[0039] and / or
[0040] The temporal forgetting function is: Among them, t∈{il i ,il i +1,il i +2,…,i-1}.
[0041] In one embodiment, the data set of the prediction results in the memory gate is in is the sample point o t (x t ,y t ) The final result predicted by the time series boosted tree model of the sample point is is the timing sample point o in the memory gate t (x t ,y t ) predicts the classification of hot encoding, t∈{il i ,il i +1,il i +2,…,i-1};
[0042] The steps of using the temporal forgetting function to perform temporal forgetting on the samples in the memory gate to obtain the temporal prediction result are as follows:
[0043]
[0044] The time series prediction result r is calculated according to the following formula (Time,i) :
[0045]
[0046] In one embodiment, the step of calculating the final classification result after time series enhancement according to the tree model prediction result and the time series prediction result includes:
[0047] Calculate the sample o according to the following formula i The corresponding sample prediction probability distribution result r after time series enhancement i :
[0048]
[0049] The sample point o is calculated according to the following formula i The final classification result after the corresponding time series enhancement is R i :
[0050]
[0051] Among them, α is the set timing weight coefficient.
[0052] In one embodiment, when the initial tree model is a single tree model, the sample prediction probability distribution result r after time series enhancement is calculated according to the following formula: i :
[0053]
[0054] Among them, α Time is a timing correction factor set based on a single tree model; or
[0055] When the initial tree model is a Bagging ensemble tree model, the sample prediction probability distribution result after time series enhancement corresponding to the Bagging ensemble tree model of C trees is r i , the sample prediction probability distribution result r after time series enhancement is calculated according to the following formula:
[0056]
[0057] Among them, r (c,i) Based on the sample point o i The sample prediction probability distribution result corresponding to the c-th tree model; or
[0058] When the initial tree model is a Boosting ensemble tree model, based on the sample point o i The K-classified Boosting ensemble tree model corresponds to the sample prediction probability distribution after time series enhancement is r i , and its calculation formula is:
[0059]
[0060] Among them, α cThe temporal weight coefficient set for the Boosting ensemble tree model.
[0061] When using the above-mentioned time series enhanced tree model classification algorithm, first use the training data to obtain the initial tree model, then use the verification data to obtain the tree model prediction result and Gini index predicted by the initial vertical tree model; use the Gini index to obtain the weight coefficient of the tree model prediction result; by setting the maximum memory length, use the weight coefficient, Gini index and maximum memory length to obtain the actual memory length; use the actual memory length to set the time series forgetting function; use the time series forgetting function to obtain the time series prediction result after time series forgetting; fuse the tree model prediction result and the time series memory of the time series prediction result to obtain the final classification result after time series enhancement. Therefore, the above-mentioned time series enhanced tree model classification algorithm can realize time series information fusion in the classification result prediction process, that is, adaptive compensation of the tree model time series information, which improves the prediction accuracy and the utilization rate of data information. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Moreover, the same reference numerals are used throughout the drawings to represent the same components. In the drawings:
[0063] Figure 1 Schematic diagram of the flow of the temporal enhanced tree model classification algorithm in a preferred embodiment of the present invention. DETAILED DESCRIPTION
[0064] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Preferred embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.
[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0066] When describing positional relationships, unless otherwise specified, when an element is referred to as being "on" another element, it can be directly on the other element or there can be an intermediate element. It is also understood that when an element is referred to as being "between" two elements, it can be the only one between the two elements, or there can be one or more intermediate elements.
[0067] In the case of using “including”, “having”, and “comprising” described herein, another component may be added unless a clear limiting term such as “only”, “consisting of”, etc. is used. Unless mentioned otherwise, a term in the singular form may include a plural form and should not be understood as being one in number.
[0068] See also Figure 1 The temporal enhanced tree model classification algorithm in the preferred embodiment of the present invention includes steps S10 to S90.
[0069] Step S10, obtaining training data and verification data.
[0070] In practical applications, the original time series data is collected through sensors, etc., and the time series data obtained after data processing (elimination of missing values and outliers, data denoising, sliding window processing, feature extraction, etc.) is used as training data and verification data.
[0071] Step S20, using a tree model algorithm to train the training data to obtain an initial tree model.
[0072] The tree model algorithm may be a common tree model algorithm such as a decision tree algorithm, a random forest algorithm, etc. The initial tree model may be a single tree model or an integrated tree model.
[0073] Step S30, input the verification data into the initial tree model, output the tree model prediction result, and obtain the Gini index of the output node corresponding to the tree model prediction result.
[0074] Specifically, the sample set of verification data is D = {(x 1 ,y 1 ),(x 2 ,y 2 ),…,(x N ,y N )},in is the feature vector, n is the number of features, y i ∈{1,2,…,K} is the classification label, K is the total number of classifications, y i ∈{1,2,…,K}, N is the sample size, and the sample point (x i ,y i ) is o i The tree model prediction result corresponding to the validation data is VTree = {r (Tree,1) ,r (Tree,2) ,…r (Tree,N)}, sample point o i The output model tree prediction results correspond to in, is the sample point o i The probability distribution results of k categories in the corresponding tree model prediction results, k = 1, 2, ...K.
[0075] Specifically in one embodiment, when the initial tree model is a single tree model or a bagging ensemble tree model, the following formula is used to calculate:
[0076]
[0077] In the above formula, is the predicted sample point o i The number of samples corresponding to the kth class of the corresponding leaf node.
[0078] Specifically in another embodiment, when the initial tree model is a Boosting ensemble tree model, the sample points corresponding to the Boosting ensemble tree model with K categories and C trees are i The probability distribution of Let r (Tree,i) =P i ;
[0079] Calculate the sample point o according to the following Softmax function formula i The probability distribution results of
[0080]
[0081] Calculate the original score logit corresponding to category k according to the following formula k :
[0082]
[0083] Among them, T k,c (o i ) is the sample point o i The output data of the c-th tree model corresponding to category k.
[0084] Specifically, the Gini index Gini (o i ):
[0085]
[0086] Step S40, calculating the weight coefficient of the tree model prediction result according to the Gini index.
[0087] Among them, the purity of the leaf node corresponding to the tree model prediction result can represent the tree model prediction confidence of the sample point, so the leaf node purity is used to represent the weight coefficient of the tree model prediction result, and the leaf node purity can be calculated by the Gini index.
[0088] Specifically, according to the following formula:
[0089] If and only if When Gini (o i ) to obtain the maximum value To determine the weight coefficient of the tree model prediction results
[0090] More specifically, the Cauchy inequality is:
[0091]
[0092] Rearranging the inequality gives:
[0093] The above Cauchy inequality combined with the calculation formula of the Gini index gives the following formula for determining the weight coefficient: The formula is:
[0094]
[0095] Therefore, the maximum value is proved by the above Cauchy inequality. The main purpose is to configure the coefficients to ensure that the time series weights of the initial tree model and the time series model are both greater than 0 and the sum of the initial weights is 1.
[0096] Step S50, setting the maximum memory length of the maximum memory gate length limit sequence.
[0097] It should be noted that the memory gate is a key means to control the length of temporal memory. Its essence is a set of historical sample points, and its length indicates the length of the algorithm's memory of temporal information.
[0098] Define the memory gate as Among them l i represents the number of memory gate samples (i.e., the actual memory length), the actual memory length l i By predicting sample point o i The purity of the corresponding leaf node Gini(o i ) and the maximum memory length l M Jointly determined.
[0099] Step S60, calculating the actual memory length according to the maximum memory length, the Gini index and the weight coefficient of the tree model prediction result.
[0100] Specifically, step S60 includes:
[0101] Set the sample point o i The memory length L i As shown in the following formula:
[0102]
[0103] The memory length L is calculated using the following formula: i Correction is made to obtain the actual memory length l i :
[0104]
[0105] Among them, l M The maximum memory length is set.
[0106] Since the memory length L i It reflects that the larger the weight of the tree model, the shorter its memory length, so the memory length L i Defined as the above formula. Considering the limited number of samples in the initial prediction, the set memory length is corrected to obtain the actual memory length to avoid the problem of insufficient initial prediction sample points and memory length greater than the number of historical samples.
[0107] Step S70, setting the temporal forgetting function according to the actual memory length.
[0108] Temporal forgetting function f (t,i) Satisfy the conditions shown in the following formula:
[0109]
[0110] The temporal forgetting function should satisfy the gradual increase of coefficients, reflecting that the closer to the sample point, the better the memory effect, and the sum of the coefficients must be 1. That is, the purpose of the first line of the formula in the above formula is to ensure that the sum of the temporal weights of all samples in the memory gate is 1 to reduce the impact of the coefficient on the weight. The second line of the formula is to ensure that the closer the temporal information is to the sample point, the greater the weight.
[0111] Specifically, the temporal forgetting function is:
[0112] Among them, t∈{il i ,il i +1,il i +2,…,i-1}.
[0113] Of course, in some other embodiments, the temporal forgetting function may also be: t∈{il i ,il i +1,…,il i}.
[0114] Of course, in some other embodiments, the forgetting function may also be: or Among them, t∈{il i ,il i +1,…,il i}, v is the speed at which the control series decays.
[0115] Of course, in some other embodiments, the temporal forgetting function can also be: Where t∈{il i ,il i +1,…,il i}.
[0116] Step S80, using a temporal forgetting function to perform temporal forgetting on the samples in the memory gate to obtain a temporal prediction result.
[0117] Suppose the data set of the prediction results in the memory gate is in is the sample point o t (x t ,y t ) The classification probability distribution result after time series enhancement corresponding to is the timing sample point o in the memory gate t (x t ,y t ) predicts the classification of hot encoding, t∈{il i ,il i +1,il i +2,…,i-1};
[0118] Step S80 includes the steps of:
[0119] The timing sample point o in the memory gate is calculated according to the following formula t (x t ,y t ) Predict the hot encoding of the classification
[0120]
[0121] The time series prediction result r is calculated according to the following formula (Time,i) :
[0122]
[0123] In this way, by executing the above two steps, the temporal forgetting of the samples in the memory gate is achieved, and the temporal prediction result is obtained.
[0124] Step S90, calculating the final classification result after time series enhancement based on the tree model prediction result and the time series prediction result.
[0125] Specifically, step S90 includes the steps of:
[0126] The sample point o is calculated according to the following formula i The corresponding sample prediction probability distribution result r after time series enhancement i :
[0127]
[0128] The sample point o is calculated according to the following formula i The final classification result after the corresponding time series enhancement is R i :
[0129]
[0130] Among them, α is the set timing weight coefficient.
[0131] Thus, the sample point o i The final classification result after the corresponding time series enhancement is r i That is, each sample point first obtains the probability distribution for all classifications, and then selects a classification with the largest probability component from these probability distributions as the final classification result of the sample point, so as to obtain the final classification category.
[0132] By executing steps S10 to S30, the tree model prediction result of the initial tree model prediction is obtained; by executing step S40, the weight coefficient of the tree model prediction result is obtained; by executing step S50, the maximum memory length is set, and by executing step S60, the actual memory length is obtained; by executing step S70, the temporal forgetting function of the memory gate is set; by executing step S80, the temporal prediction result after temporal forgetting can be obtained; by executing step S90, the temporal memory fusion of the tree model prediction result and the temporal prediction result can be realized to obtain the final classification result after temporal enhancement. Therefore, the above-mentioned temporal enhancement tree model classification algorithm can realize temporal information fusion in the classification result prediction process, that is, adaptive compensation of the tree model temporal information, which improves the prediction accuracy and data information utilization rate during the prediction of temporal data.
[0133] In some embodiments, when the initial tree model is a single tree model, the sample prediction probability distribution result r after time series enhancement is calculated according to the following formula: i :
[0134]
[0135] Among them, α Time The timing correction factor is set based on a single tree model.
[0136] Thus, the above-mentioned temporal enhanced tree model classification algorithm is suitable for application scenarios of a single tree model.
[0137] In some embodiments, when the initial tree model is a bagging ensemble tree model, the sample prediction probability distribution result after time series enhancement corresponding to the bagging ensemble tree model of C trees is r i , calculate the sample prediction probability distribution result r after time series enhancement according to the following formula i :
[0138]
[0139] Among them, r (c,i) Based on the sample point o i The c-th tree corresponds to the probability distribution result after the output time series enhancement.
[0140] For the Bagging ensemble tree model, due to the randomness of sampling and feature selection, the probability distribution of the time series enhancement corresponding to the Bagging ensemble tree model with C trees is obtained by averaging, which is r i Therefore, the above-mentioned time series enhanced tree model classification algorithm is suitable for the application of Bagging ensemble tree model.
[0141] In some embodiments, when the initial tree model is a Boosting ensemble tree model, based on the sample points o i The K-classified Boosting ensemble tree model corresponds to the sample prediction probability distribution after time series enhancement is r i , and its calculation formula is:
[0142]
[0143] Among them, α c is the timing correction coefficient based on the Boosting ensemble tree model.
[0144] The calculation formula for the above sample prediction probability distribution result r is derived as follows:
[0145] When the initial tree model is a Boosting ensemble tree model, considering that the Boosting ensemble tree model is often better than a single tree model in terms of prediction effect, the tree model prediction result needs to be coefficient corrected by the result of temporal forgetting, that is, multiplied by the coefficient α Tree (αTree >1), to obtain the following formula:
[0146]
[0147] And r i is the sample point o i Corresponding to the probability distribution of K categories, the final classification result after time series enhancement only takes the category corresponding to the maximum probability component. In order to reduce the number of parameters, the above formula is divided by the coefficient α Tree ,remember Get the sample prediction probability distribution result r after time series enhancement corresponding to the Boosting ensemble tree model of K classifications i The calculation formula is: Therefore, the above-mentioned time series boosted tree model classification algorithm is suitable for the application occasions of the Boosting ensemble tree model.
[0148] It should be noted that α, α c and α Time are respectively the timing correction coefficients set in different embodiments.
[0149] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0150] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A temporal enhanced tree model classification algorithm, characterized in that: Includes steps: Obtain training data and validation data; Using a tree model algorithm to train the training data to obtain an initial tree model; Input the verification data into the initial tree model, output the tree model prediction result, and obtain the Gini index of the output node corresponding to the tree model prediction result; Calculate the weight coefficient of the tree model prediction result according to the Gini index; Set the maximum memory gate length to limit the maximum memory length of the timing information; The actual memory length is calculated according to the maximum memory length, the Gini index and the weight coefficient of the tree model prediction result; Setting a temporal forgetting function according to the actual memory length; Using the temporal forgetting function to perform temporal forgetting on the samples in the memory gate to obtain a temporal prediction result; The final classification result after time series enhancement is calculated based on the tree model prediction result and the time series prediction result.
2. The temporal enhanced tree model classification algorithm according to claim 1, characterized in that: The sample set of the verification data is D = {(x1, y1), (x2, y2), ..., (x N ,y N )};in, is the feature vector, n is the number of features, y i ∈{1,2,…,K} is the classification mark, K is the total number of classifications, i=1,2,…,N, N is the sample size, and the sample point (x i ,y i ) is o i ; The tree model prediction result corresponding to the sample set of the verification data is V Tree = {r (Tree,1) ,r (Tree,2) ,…r (Tree,N) }, sample point o i The corresponding tree model prediction result is in, is the sample point o i The probability distribution result of the kth classification in the corresponding tree model prediction result, k = 1, 2, ...K.
3. The temporal enhanced tree model classification algorithm according to claim 2 is characterized in that: When the initial tree model is a single tree model or a bagging ensemble tree model, it is calculated according to the following formula In the above formula, is the predicted sample point o i The number of samples corresponding to the kth class of the corresponding leaf node; or When the initial tree model is a Boosting ensemble tree model, the sample point o corresponding to the Boosting ensemble tree model with K classifications and C trees is i The probability distribution of Let r (Tree,i) =P i ; Calculate the sample point o according to the following Softmax function formula i The probability distribution result P i (k) : Calculate the original score logit corresponding to category k according to the following formula k : Among them, T k,c (o i ) is the sample point o i The output data corresponding to the c-th tree model of category k.
4. The temporal enhanced tree model classification algorithm according to claim 3 is characterized in that: The step of obtaining the Gini index of the output node corresponding to the prediction result of the tree model is as follows: calculate the Gini index Gini(o i ):
5. The temporal enhanced tree model classification algorithm according to claim 4, characterized in that: The steps of calculating the weight coefficient of the initial tree model result according to the Gini index are: According to the following formula: If and only if When Gini (o i ) to obtain the maximum value To determine the weight coefficient of the tree model prediction results 6. The temporal enhanced tree model classification algorithm according to claim 5, characterized in that: The step of calculating the actual memory length according to the maximum memory length and the Gini index comprises the steps of: Set the sample point o i The memory length L i As shown in the following formula: The memory length L is calculated using the following formula: i Correction is made to obtain the actual memory length l i : Among them, l M The maximum memory length is set.
7. The temporal enhanced tree model classification algorithm according to claim 6, characterized in that: The temporal forgetting function f (t,i) Satisfy the conditions shown in the following formula: The temporal forgetting function is: Among them, t∈{il i ,il i +1,il i +2,…,i-1}.
8. The temporal enhanced tree model classification algorithm according to claim 7, characterized in that: Suppose the data set of the prediction results in the memory gate is in is the sample point o t (x t ,y t ) The final result predicted by the time series boosted tree model of the sample point is is the timing sample point o in the memory gate t (x t ,y t ) predicts the classification of hot encoding, t∈{il i ,il i +1,il i +2,…,i-1}; The steps of using the temporal forgetting function to perform temporal forgetting on the samples in the memory gate to obtain the temporal prediction result are as follows: The time series prediction result r is calculated according to the following formula (Time,i) :
9. The temporal enhanced tree model classification algorithm according to claim 8, characterized in that: The step of calculating the final classification result after time series enhancement according to the tree model prediction result and the time series prediction result comprises: Calculate the sample o according to the following formula i The corresponding prediction probability distribution result r after time series enhancement i : The sample point o is calculated according to the following formula i The final classification result after the corresponding time series enhancement is R i : Among them, α is the set timing weight coefficient.
10. The temporal enhanced tree model classification algorithm according to claim 9, characterized in that: When the initial tree model is a single tree model, the sample prediction probability distribution result r after time series enhancement is calculated according to the following formula: i : Among them, α Time is a timing correction factor set based on a single tree model; or When the initial tree model is a Bagging ensemble tree model, the sample prediction probability distribution result after time series enhancement corresponding to the Bagging ensemble tree model of C trees is r i , calculate the sample prediction probability distribution result r after time series enhancement according to the following formula i : Among them, r (c,i) Based on the sample point o i The sample prediction probability distribution result corresponding to the c-th tree model; or When the initial tree model is a Boosting ensemble tree model, based on the sample point o i The predicted probability distribution result after time series enhancement corresponding to the K-classified Boosting ensemble tree model is r i , and its calculation formula is: Among them, α c The temporal weight coefficient set for the Boosting ensemble tree model.