WO2025007412A1 - 多中心小样本场景下目标预测模型构建方法和预测方法 - Google Patents

多中心小样本场景下目标预测模型构建方法和预测方法 Download PDF

Info

Publication number
WO2025007412A1
WO2025007412A1 PCT/CN2023/116931 CN2023116931W WO2025007412A1 WO 2025007412 A1 WO2025007412 A1 WO 2025007412A1 CN 2023116931 W CN2023116931 W CN 2023116931W WO 2025007412 A1 WO2025007412 A1 WO 2025007412A1
Authority
WO
WIPO (PCT)
Prior art keywords
knowledge
training
prediction
prediction model
classifier
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2023/116931
Other languages
English (en)
French (fr)
Inventor
钱鹏江
王志煌
王士同
蒋亦樟
方伟
宋威
姚健
张欣
陈爱国
顾逸
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangnan University
Original Assignee
Jiangnan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangnan University filed Critical Jiangnan University
Priority to US18/412,519 priority Critical patent/US12260350B2/en
Publication of WO2025007412A1 publication Critical patent/WO2025007412A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/048—Fuzzy inferencing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2337—Non-hierarchical techniques using fuzzy logic, i.e. fuzzy clustering
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry

Definitions

  • the present invention relates to the field of computer technology, and in particular to a target prediction model construction method and a prediction method in a multi-center small sample scenario.
  • KNN K-Nearest Neighbor
  • SVM Support Vector Machine
  • NB Naive Bayes
  • LDA Latent Dirichlet Allocation
  • CART Classification and Regression tree
  • the present invention provides a target prediction model construction method and a prediction method in a multi-center small sample scenario.
  • a method for constructing a target prediction model in a multi-center small sample scenario includes the following steps:
  • S1 Input the features of the training samples in a multi-center training data set with m data centers, use multiple sub-classifiers and a historical knowledge classifier to predict the training samples respectively, and the historical knowledge classifier is a classifier that integrates the rules of all sub-classifiers; integrate all the prediction results to have a total of m+1 dimensions, each dimension represents the prediction result of a classifier, and if it is greater than 0, it is judged as a positive class, and if it is less than 0, it is a negative class;
  • P is the prediction vector
  • W is the weight vector of the sub-classifier to be trained
  • Y is the true value label of the corresponding training sample
  • ⁇ is the regularization parameter of the structural risk term
  • Pg is the consequent parameter calculated by the ridge regression method
  • I is the unit matrix.
  • the training of the first sub-classifier includes two stages: a basic stage and a learning stage.
  • the basic stage includes: obtaining initial fuzzy rule parameters c, P g , ⁇ and misclassified sample information errors by means of classic 0th order TSK, and using Gaussian kernel function as membership function;
  • c and ⁇ are antecedent parameters
  • c is the cluster center obtained by FCM clustering of the input data Data
  • ⁇ is the membership matrix U of the cluster center obtained by FCM clustering of the input data Data and c
  • the kernel width of the membership function is obtained by FCM clustering of the input data Data
  • c and ⁇ are antecedent parameters
  • c is the cluster center obtained by FCM clustering of the input data Data
  • ⁇ is the membership matrix U of the cluster center obtained by FCM clustering of the input data Data and c
  • the kernel width of the membership function is the kernel width of the membership function.
  • the learning stage includes: first performing knowledge discarding processing on c and P g , then sorting the knowledge to obtain the best knowledge combination, and finally calculating the consequent knowledge corresponding to the current fuzzy rule in combination with the input data Data to obtain the prediction result.
  • the knowledge discarding process includes: first, determining the knowledge discarding ratio r1 and the knowledge confusion feature ratio r2 , randomly selecting the rule to be subjected to knowledge discarding according to r1 , and randomly selecting the feature column lose_columns to be subjected to knowledge confusion processing according to r2 ; for each rule to be subjected to knowledge discarding, randomly determining whether knowledge forgetting or knowledge confusion is to be performed; for the rule to be subjected to knowledge forgetting, directly setting c and Pg to empty; for the rule to be subjected to knowledge confusion processing, replacing the lose_columns column of the rule with the corresponding column value of other rules.
  • the knowledge organization includes: taking the discarded c and P g and the misclassified sample information errors as candidate knowledge, and selecting the best knowledge combination therefrom as preliminary knowledge for the current stage, wherein the misclassified sample information errors includes sample features and sample labels corresponding to c and P g respectively.
  • Xg is the input vector of the training sample after Gaussian kernel mapping
  • Pg -history is the historical post-process parameter migration knowledge.
  • the first sub-classifier is used as a guarantee for the rapid training of subsequent sub-classifiers, and the first sub-classifier should be trained in a data center with relatively sufficient data volume.
  • the training of the first sub-classifier includes two stages: a basic stage and a learning stage.
  • the first-order form and the zero-order form of the classic 0-order TSK fuzzy system are used to obtain the initial fuzzy rule parameters c, P g , ⁇ and the error classification sample information errors, and the Gaussian kernel function is used as the membership function.
  • c and ⁇ are antecedent parameters
  • c is the cluster center obtained by the input data Data using the fuzzy C-means algorithm (fuzzy c-means, FCM clustering)
  • ⁇ is the membership matrix U of the Data to the cluster center obtained by the input data Data, FCM clustering
  • the membership function kernel width obtained by c are antecedent parameters
  • c is the cluster center obtained by the input data Data using the fuzzy C-means algorithm (fuzzy c-means, FCM clustering)
  • ⁇ is the membership matrix U of the Data to the cluster center obtained by the input data Data, FCM clustering
  • the membership function kernel width obtained by c.
  • the learning phase includes the following steps:
  • the knowledge discarding mechanism simulates the forgetting characteristics of the human brain, including knowledge forgetting and knowledge confusion. Corresponding to the two situations of complete knowledge forgetting and partial knowledge feature forgetting. First, determine the knowledge discarding ratio r 1 and the knowledge confusion feature ratio r 2 , randomly select the rule to be discarded according to r 1 , and randomly select the feature column lose_columns to be processed for knowledge confusion according to r 2. For each rule to be discarded, randomly determine whether to forget knowledge or be confused. For the rule to be forgotten, just leave c and P g empty; for the rule to be confused, replace the lose_columns column of the rule with the corresponding column value of other rules to achieve the effect of knowledge confusion.
  • the discarded c and P g and the misclassified sample information errors are taken as candidate knowledge, and the best knowledge combination is selected as the preliminary knowledge of the current stage.
  • the misclassified sample information errors includes sample features and sample labels corresponding to c and P g respectively.
  • the distance calculation involved in knowledge arrangement is the distance between the centers of the center set composed of the feature columns of the antecedent knowledge c and errors. The selection of knowledge follows two principles: high reliability and low similarity.
  • the number of training samples in the neighborhood of each candidate center needs to be calculated, and then the value is mapped to [0,1].
  • the more training samples in the neighborhood the higher the credibility of the candidate center.
  • the neighborhood radius is the average of the sum of the Euclidean distances between all candidate centers.
  • the similarity between each candidate center and the representative center set is estimated by calculating the Euclidean distance between the candidate center and the nearest center in the representative center set, and an attempt is made to ensure that the similarity between the selected centers is low.
  • the similarity value also needs to be mapped to [0,1].
  • the importance of each center is reflected by calculating the sum of the mapped reliability and similarity indicators, and the final best center combination is selected based on this. Finally, these centers and the subsequent knowledge corresponding to these centers are selected as the best knowledge combination.
  • the membership function kernel width ⁇ can be calculated by combining the input data Data and the preliminary knowledge c.
  • the empirical risk term, the structural risk term and the historical consequent parameter migration knowledge regularization term are used as the objective function to quickly calculate the consequent knowledge corresponding to the current fuzzy rule.
  • the objective function expression and the consequent parameter solution formula are as follows:
  • Xg is the input vector of the training sample after Gaussian kernel mapping
  • Pg -history is the parameter migration knowledge of the historical post-process. Therefore, with the help of historical knowledge and knowledge of misclassified samples, further generalization learning of knowledge is achieved, and more robust new knowledge is continuously obtained.
  • the collaborative training process of multiple sub-classifiers is shown in FIG3.
  • the generated fuzzy rules are updated and added to the historical fuzzy rule base for continued use by subsequent sub-classifiers.
  • the basic stage training is no longer required. Instead, the knowledge of the current historical fuzzy rule base is directly used to predict the data samples of the center.
  • the training of the learning stage can be carried out like the first sub-classifier, which greatly saves training time and makes better use of historical knowledge, avoiding the situation where good knowledge cannot be obtained in the basic stage in a small sample scenario.
  • the present invention also provides a target prediction method in a multi-center small sample scenario, the steps comprising:
  • the prediction model obtains a prediction result of the current target according to the features.
  • the target prediction model construction method and prediction method in a multi-center small sample scenario of the present invention can be applied to scenarios such as traffic condition prediction and disease auxiliary diagnosis.
  • each hospital can develop a more reasonable prediction model under the guidance of the fuzzy rules of other hospitals, which becomes an auxiliary means for doctors' diagnosis, greatly reducing the workload of doctors, improving the efficiency of patients' medical treatment, and making full use of the data of different hospitals.
  • This application uses the personal-key-indicators-of-heart-disease dataset on the Kaggle website (https://www.kaggle.com/datasets/kamilpytlak/personal-key-indicators-of-heart-disease) as the experimental dataset. Due to the excessive amount of data in the original dataset, in order to simulate the small sample node scenario of a multi-center dataset, the experiment selected some samples from it for testing. First, according to the ethnic feature column in the dataset, whites, blacks, and Asians were taken as the three centers to produce the dataset. The feature column information of the selected data samples is complete, and the character feature columns are replaced by discrete numbers for subsequent calculations. Finally, each selected sample includes 16 feature columns and 1 label column.
  • the positive class represents sample data with heart disease
  • the negative class represents sample data without heart disease
  • S1 Data preprocessing. According to race, samples with complete feature information are selected from three centers: white, black and Asian to make data sets, and character feature columns are replaced with discrete numbers for subsequent calculations. Each sample finally selected includes 16 feature columns and 1 label column, where the label column is used to calculate the weights of each sub-classifier or to verify the accuracy of the target prediction model construction method in the multi-center small sample scenario of this application.
  • the training set and test set were randomly divided for each center in a ratio of 4:1, and the data features were normalized to map the features to the range of [-1,1].
  • S2 Create a multi-center dataset. According to the processed data, save the training set and test set of each center in .mat format, and also save the training set of all centers and the test set of all centers in .mat format, so as to finally train the weight of each sub-classifier.
  • S3 Model training and parameter range adjustment. Since the network will have different effect deviations for a specific data set, grid optimization is required. The main purpose is to find the best parameters for a specific network for a specific data set. Use a 5-fold cross-validation method to find the optimal parameters from the parameter optimization range. By continuously updating the current optimal parameters until the iteration ends, the final optimal hyperparameters are obtained. In this process, the model results are observed through multiple experiments, combined with the performance of the model on the training set and the test set, and the optimal parameters are determined. Determine whether the model has achieved the best fit and gradually determine the optimal optimization range of the model parameters. In the above data set, the relevant parameters of the algorithm during optimization are as follows:
  • S6 Evaluation indicators. The test results of the test set have been obtained through the above steps. Multiple evaluation indicators are calculated using the formula. The evaluation indicators used in this experiment are accuracy, precision, recall, F1score and AUC.
  • This embodiment compares the target prediction model construction method in the multi-center small sample scenario provided by the present application with the existing classic traditional machine learning algorithms, including: Linear Discriminant Analysis (LDA), K-Nearest Neighbor (KNN), Naive Bayes ( Bayes, NB), classification and regression tree algorithm (classification and regression tree, The first-order and zero-order forms of CART, support vector machines (SVM) and TSK fuzzy system (TSK-FS).
  • LDA Linear Discriminant Analysis
  • KNN K-Nearest Neighbor
  • Naive Bayes Bayes, NB
  • classification and regression tree algorithm classification and regression tree
  • the first-order and zero-order forms of CART support vector machines (SVM) and TSK fuzzy system (TSK-FS).
  • SVM support vector machines
  • TSK-FS TSK fuzzy system
  • AUC Absolute Under Curve
  • the AUC value is used as the evaluation criterion of the model because in many cases the ROC curve cannot clearly indicate which classifier is more effective, and as a numerical value, the classifier with a larger AUC is more effective.
  • the target prediction model construction method in the multi-center small sample scenario provided by this application has better performance than other algorithms, and is comparable to other traditional algorithms in terms of recall rate and F1 score. Therefore, from the perspective of classification performance, the target prediction model construction method in the multi-center small sample scenario provided by this application has better performance than other traditional algorithms.
  • 0-TSK-FS has a lower rule complexity than the target prediction model construction method for multi-center small sample scenarios provided in this application
  • the classification performance of the target prediction model construction method for multi-center small sample scenarios provided in this application is obviously much better than 0-TSK-FS.
  • 1-TSK-FS is not as good as the target prediction model construction method for multi-center small sample scenarios provided in this application in terms of classification performance and rule complexity.
  • the classification performance and interpretability of the target prediction model construction method for multi-center small sample scenarios provided in this application are better than these traditional algorithms.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Business, Economics & Management (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Automation & Control Theory (AREA)
  • Fuzzy Systems (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Human Resources & Organizations (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明涉及一种多中心小样本场景下目标预测模型构建方法和预测方法。通过结合迁移学习的思想直接用已训练节点的知识对新节点的训练集进行预测,借助预测错误样本体现新节点与已训练节点的差异,将其作为补充知识,以此快速获取新节点的模型知识,避免每次都对新节点从头训练;最后使用岭回归方法实现增量式子分类器的并行集成,大大节省了部署时间和成本。通过共享历史知识和知识丢弃机制保证模型的泛化性,在足够的共享知识的支持下,对于小样本量的节点也可以取得较好的分类效果。基于具有高度可解释性的零阶TSK模糊系统进行改进,所有的模型参数都可以通过规则的形式体现,有着高度的语义透明性,在实际应用中更能帮助用户分析和理解数据。

Description

多中心小样本场景下目标预测模型构建方法和预测方法 技术领域
本发明涉及计算机技术领域,尤其是指一种多中心小样本场景下目标预测模型构建方法和预测方法。
背景技术
随着计算机算力的提升以及大数据时代的到来,机器学习成为数据分析的有力工具,对人类生活质量的提高做出了突出贡献。然而,由于数据的隐私性,不同机构、部门之间存在数据壁垒,导致海量数据难以实现共享,多中心数据的部署和联合训练成了难题。为此,谷歌提出联邦学习(federated learning)技术,通过在本地阶段对各自的数据进行存储和训练,仅与中心服务器交互模型更新,有效保障了用户的隐私安全。目前,虽然该领域已经存在诸多算法,但大部分算法基于深度学习网络提出,目前的深度学习方法参数多,模型大,数据泄露风险高,部署困难,虽然分类精度已经可以达到十分理想的程度,但是大多都是“黑盒模型”,缺乏语义可解释性,在高风险领域达不到安全性的要求,对于要求高度安全性和可解释性的机构并不适用。而传统机器学习算法,如邻近算法(K-NearestNeighbor,KNN)、支持向量机(Support Vector Machine,SVM)、朴素贝叶斯法(Naive Bayes,NB)、隐含狄利克雷分布(Latent Dirichlet Allocation,LDA)、分类与回归树(Classification and Regression tree,CART)等,对于多中心数据也容易因数据漂移和小样本问题而导致准确率不高,没有深度学习分类效果好,也无法保证多中心数据隐私问题。
发明内容
为解决上述技术问题,本发明提供了一种多中心小样本场景下目标预测模型构建方法和预测方法。
一种多中心小样本场景下目标预测模型构建方法,包括以下步骤:
S1:在具有m个数据中心的多中心训练数据集中输入训练样本的特征,使用多个子分类器以及历史知识分类器分别对训练样本进行预测,所述历史知识分类器为融合了所有子分类器规则的分类器;整合所有预测结果共有m+1维,每个维度代表一个分类器的预测结果,大于0则判断为正类,小于0则为负类;
S2:将预测结果向量归一化到[-1,1]之间;
S3:以归一化后的预测向量为预测特征,结合训练样本对应的训练标签,使用岭回归方法训练每个子分类器的权重;
S4:结合预测向量和训练得到的分类器权重得到综合预测结果;
S5:对综合预测结果进行分类,得到与所述训练标签相同的结果。
优选的,所述权重训练的目标函数以及权重参数的求解公式为:

Pg=(λI+PTP)-1PTY,
其中,P为预测向量,W为要训练的子分类器权重向量,Y为对应训练样本的真值标签,λ为结构风险项的正则化参数,Pg为通过岭回归方法计算得到的后件参数,I为单位矩阵。
优选的,第一子分类器的训练包括两个阶段:基础阶段和学习阶段。
优选的,所述基础阶段包括:借助经典0阶TSK来获取初始模糊规则参数c、Pg、δ以及错误分类样本信息errors,使用高斯核函数作为隶属度函数; 其中,c和δ为前件参数,c是输入数据Data借助FCM聚类得到的聚类中心,δ是根据输入数据Data、FCM聚类得到的Data对聚类中心的隶属度矩阵U以及c得到的隶属度函数核宽度。
优选的,所述学习阶段包括:先对c和Pg执行知识丢弃处理,然后进行知识整理得到最佳知识组合,最后结合输入数据Data计算出当前模糊规则对应的后件知识,得到预测结果。
优选的,所述知识丢弃处理包括:首先,确定知识丢弃比例r1和知识混乱特征比例r2,根据r1随机选择出要进行知识丢弃的规则,根据r2随机选择要进行知识混乱处理的特征列lose_columns;对于每一条要进行知识丢弃的规则,随机确定要进行知识遗忘或知识混乱;要进行知识遗忘的规则直接将c和Pg置空;对于要进行知识混乱处理的规则,将该规则的lose_columns列以其他规则的相对应列值进行替换。
优选的,所述知识整理包括:将丢弃处理后的c和Pg以及错误分类样本信息errors一起作为候选知识,从中选出最佳知识组合作为当前阶段的预备知识,所述错误分类样本信息errors包括分别对应c和Pg的样本特征和样本标签。
优选的,结合输入数据Data计算出当前模糊规则对应的后件知识,得到预测结果具体包括:结合输入数据Data和预备知识中的c计算出隶属度函数核宽度δ;利用前件参数c和δ以及对应后件参数的历史迁移知识,通过岭回归的方法,将经验风险项、结构风险项以及历史后件参数迁移知识正则化项作为目标函数计算出当前模糊规则对应的后件知识,目标函数表达式以及后件参数求解公式如下:

其中,Xg是高斯核映射后训练样本的输入向量,Pg-history历史后件参数迁移知识。
优选的,每个子分类器训练之后都将产生的模糊规则更新加入到历史模糊规则库中,以便后续子分类器的继续使用。
一种多中心小样本场景下目标预测方法,步骤包括:
S1:采用如上所述的多中心小样本场景下目标预测模型构建方法构建预测模型;
S2:在所述预测模型中输入当前目标中与训练样本特征对应的特征;
S3:所述预测模型根据所述特征得出该当前目标的预测结果。
本发明的上述技术方案相比现有技术具有以下优点:
1、本发明所述的多中心小样本场景下目标预测模型构建方法结合迁移学习的思想直接用已训练节点的知识来对新节点的训练集进行预测,借助预测错误样本来体现新节点与已训练节点的差异,将其作为补充知识,以此快速获取新节点的模型知识,避免每次都对新节点从头训练,最后使用岭回归方法即可实现增量式子分类器的并行集成,大大节省了部署时间和成本。
2、本发明所述的多中心小样本场景下目标预测模型构建方法通过共享历史知识和知识丢弃机制来保证模型的泛化性,在足够的共享知识的支持下,对于小样本量的节点也可以取得较好的分类效果。
3、所述的多中心小样本场景下目标预测模型构建方法基于具有高度可解释性的零阶TSK模糊系统进行改进,所有的模型参数都可以通过规则的形式体现,有着高度的语义透明性,在实际应用中更能帮助用户分析和理解数据。
4、本发明所述的多中心小样本场景下目标预测模型构建方法很好地保证了不同节点间的样本不会相互泄露,并且在实验中表现出了更好的分类性能。
附图说明
为了使本发明的内容更容易被清楚的理解,下面根据本发明的具体实施例并结合附图,对本发明作进一步详细的说明。
图1是本发明多中心小样本场景下目标预测模型构建方法整体模型架构。
图2是本发明多中心小样本场景下目标预测模型构建方法第一子分类器训练细节。
图3是本发明多中心小样本场景下目标预测模型构建方法子分类器的协同训练过程。
图4是本发明多中心小样本场景下目标预测模型构建方法实验流程图。
具体实施方式
下面结合附图和具体实施例对本发明作进一步说明,以使本领域的技术人员可以更好地理解本发明并能予以实施,但所举实施例不作为对本发明的限定。
如图1所示,本发明提供一种多中心小样本场景下目标预测模型构建方法,包括以下步骤:
S1:在具有m个数据中心的多中心训练数据集中输入训练样本的特征,使用多个子分类器以及历史知识分类器分别对训练样本进行预测,所述历史知识分类器为融合了所有子分类器规则的分类器;整合所有预测结果共有m+1维,每个维度代表一个分类器的预测结果,大于0则判断为正类,小于0则为负类。
S2:将预测结果向量归一化到[-1,1]之间,如此操作可保证量纲统一。
S3:以归一化后的预测向量为预测特征,结合训练样本对应的训练标签,使用岭回归方法训练每个子分类器的权重,权重训练的目标函数以及权重参数的求解公式为:

Pg=(λI+PTP)-1PTY,
其中,P为预测向量,W为要训练的子分类器权重向量,Y为对应训练样本的真值标签,λ为结构风险项的正则化参数,Pg为通过岭回归方法计算得到的后件参数,I为单位矩阵。通过以上述公式利用岭回归的方法可以迅速得到权重W的解。
S4:结合预测向量和训练得到的分类器权重计算综合预测结果。
S5:对于预测结果进行分类,若大于0则为正类,反之为负类。
在一个具体的实施例中,第一子分类器作为后续子分类器快速训练的保障,第一子分类器应该选择数据量相对充足的数据中心进行训练。如图2所示,第一子分类器的训练包括两个阶段:基础阶段和学习阶段。
在基础阶段中,借助经典0阶TSK模糊系统(Takagi-Sugeno-Kang,TSK)的一阶形式和零阶形式来获取初始模糊规则参数c、Pg、δ以及错误分类样本信息errors,使用高斯核函数作为隶属度函数。其中,c和δ为前件参数,c是输入数据Data借助模糊C均值算法(fuzzy c-means,FCM聚类)得到的聚类中心,δ是根据输入数据Data、FCM聚类得到的Data对聚类中心的隶属度矩阵U以及c得到的隶属度函数核宽度。
在学习阶段中,包括以下步骤:
S1:先对c和Pg执行知识丢弃处理,
S2:然后进行知识整理得到最佳知识组合,
S3:最后结合输入数据Data计算出当前模糊规则对应的后件知识,得到预测结果。
在S1中,知识丢弃机制模拟人脑的遗忘特性,包括知识遗忘和知识混乱, 对应于知识完全遗忘和知识部分特征遗忘两种情况。首先,确定知识丢弃比例r1和知识混乱特征比例r2,根据r1随机选择出要进行知识丢弃的规则,根据r2随机选择要进行知识混乱处理的特征列lose_columns。对于每一条要进行知识丢弃的规则,随机确定要进行知识遗忘或知识混乱。要进行知识遗忘的规则直接将c和Pg置空即可;对于要进行知识混乱处理的规则,将该规则的lose_columns列以其他规则的相对应列值进行替换,达到知识错乱的效果。
在S2中,将丢弃处理后的c和Pg以及错误分类样本信息errors一起作为候选知识,从中选出最佳知识组合作为当前阶段的预备知识,错误分类样本信息errors包括分别对应c和Pg的样本特征和样本标签。知识整理涉及的距离计算是前件知识c以及errors的特征列构成的中心集中的中心的距离。知识的选择遵循两个原则:高可靠性和低相似性。
首先,为了衡量候选中心的可信度,需要计算每个候选中心邻域内的训练样本数,然后将值映射到[0,1]。邻域中的训练样本越多,候选中心的可信度就越高。邻域半径是所有候选中心之间的欧几里得距离之和的平均值。此外,通过计算候选中心与代表中心集中最近的中心之间的欧几里得距离来估计每个候选中心和代表中心集之间的相似度,并试图确保所选中心之间的相似度较低。同样,相似性值也需要映射到[0,1]。通过计算映射后的可靠性和相似性的指标之和来反映每个中心的重要程度,以此为依据选出最终的最佳中心组合,最后将这些中心以及这些中心相对应的后件知识作为选中的最佳知识组合。
在S3中,结合输入数据Data和预备知识中的c,即可计算出隶属度函数核宽度δ。此时,利用前件参数c和δ以及对应后件参数的历史迁移知识,通过岭回归的方法,将经验风险项、结构风险项以及历史后件参数迁移知识正则化项作为目标函数即可快速计算出当前模糊规则对应的后件知识,目标函数表达式以及后件参数求解公式如下:

其中,Xg是高斯核映射后训练样本的输入向量,Pg-history历史后件参数迁移知识。由此,借助历史知识和错误分类样本知识实现了知识的进一步泛化学习,不断得到更加稳健的新知识。
在一个具体的实施例中,多个子分类器的协同训练过程如图3所示,每个子分类器训练之后都将产生的模糊规则更新加入到历史模糊规则库中,以便后续子分类器的继续使用。对于第一个子分类器之外的其他子分类器,不再需要基础阶段的训练,而是直接使用当前历史模糊规则库的知识对本中心数据样本进行预测,得到错误分类样本信息errors后,便可以像第一个子分类器一样进行学习阶段的训练,大大节省了训练时间,也更好地利用了历史知识,避免小样本场景下基础阶段无法得到较好知识的情况。
本发明还提供一种多中心小样本场景下目标预测方法,步骤包括:
S1:采用如上所述的多中心小样本场景下目标预测模型构建方法构建预测模型;
S2:在所述预测模型中输入当前目标中与训练样本特征对应的特征;
S3:所述预测模型根据所述特征得出该当前目标的预测结果。
本发明的多中心小样本场景下目标预测模型构建方法和预测方法可以应用于交通情况预测、疾病辅助诊断等场景。
当交通堵塞时,如果没有提前派遣足够的交警前往疏通交通,导致交通事故和违章事件的发生,给人们的生命财产安全带来严重威胁。具体的,进行交通情况预测时,以不同城市作为数据中心,各数据中心的交通堵塞时间,堵塞地点等由于各种原因会有一定的不同,但又有相似的趋势。各城市分别 采集其多年内的交通数据作为训练特征共同采用本申请所述的多中心小样本场景下目标预测模型构建方法进行拥堵情况预测模型构建,实现不同的数据中心的交通数据在不对其他数据中心泄露的情况下实现抽象模糊规则的共享,各中心可将当前交通数据作为特征输入至该拥堵情况预测模型,得到当前拥堵情况的预测结果。该预测结果借鉴其他中心的堵塞特征趋势,从而更加合理地判断哪些时段更容易出现堵塞高峰,以便及时地派遣交警队员前往保障交通,保证人们正常通行和行车安全。
在医疗领域,医生通常无法提供大量的标注数据供人工智能模型训练,大部分情况下只能得到少量的有标注样本,而小样本训练得到的预测结果通常是比较差的。对于不同医院的医疗数据,虽然由于地区等原因数据分布会有所不同,但还是存在相似的趋势,对彼此模型的训练有着很大的指导意义。然而,为了患者的隐私考虑,医疗数据一般不能向外公开,因此虽然多个医院可以拥有大量的数据但彼此却不能通用。本申请所述的多中心小样本场景下目标预测模型构建方法可以提取出每个医院的抽象的模糊规则,在保证患者隐私的前提下,每个医院都可以在其他医院模糊规则的指导下开发出更加合理的预测模型,成为医生诊断的辅助手段,大大减少了医生的工作量,提高了患者的就医效率,使得不同医院的数据都得到了充分的利用。
本申请以Kaggle网站上的personal-key-indicators-of-heart-disease数据集(https://www.kaggle.com/datasets/kamilpytlak/personal-key-indicators-of-heart-disease)作为实验数据集。由于原数据集数据量过多,为了模拟多中心数据集的小样本节点场景,实验从中选取了部分样本进行试验。首先,根据数据集中的人种特征列,分别取白种人,黑种人以及亚洲人作为三个中心来制作数据集,选取的数据样本特征列信息完整,并将字符型特征列以离散数字代替以便后续计算,最终选中的每个样本都包括16个特征列和1个标签列。实验中,以4:1的比例划分训练集和测试集,并采用五折交叉验证确定参数,对数据特征进行归一化操作,将特征映射到[-1,1]的范围内。下表记录了数据集 的具体划分情况:
其中,正类代表患了心脏病的样本数据,负类代表未患心脏病的样本数据。
具体实验步骤如图4所示,包括:
S1:数据预处理。根据人种,分别从白种人,黑种人以及亚洲人三个中心中选择特征信息完整的样本用于制作数据集,并将字符型特征列以离散数字代替以便后续计算,最终选中的每个样本都包括16个特征列和1个标签列,其中,标签列用于计算各子分类器的权重或验证本申请多中心小样本场景下目标预测模型构建方法预测数据的准确性。实验中,为每个中心以4:1的比例随机划分训练集和测试集,并对数据特征进行归一化操作,将特征映射到[-1,1]的范围内。
S2:制作多中心数据集。根据处理好的数据,把每个中心的训练集和测试集分别用.mat格式保存,并额外将所有中心的训练集和所有中心的测试集也用.mat格式保存起来,以便最后训练每个子分类器的权重。
S3:模型训练与参数范围调整。由于网络针对特定的数据集会有不同效果的偏差,所以要进行网格寻优,主要作用是特定网络针对特定的数据集取到最佳参数。以五折交叉验证的方式从参数寻优范围中寻找最优参数,通过不断更新当前最优参数直至迭代结束,得到最终的最优超参数。在这一过程中,通过多次实验观测模型结果,结合模型在训练集和测试集上的表现,判 断模型是否达到最佳拟合,逐步确定模型参数的最佳寻优范围。在上述数据集中,该算法在寻优时的相关参数具体如下:
S4:再次训练。使用网格寻优查找到的最优超参数来重新进行模型训练。
S5:模型测试。利用测试集对已经训练好的模型进行测试,观察测试结果是否达到预期。
S6:评价指标。通过上述步骤已经得到测试集的测试结果,利用公式计算多个评价指标,本次实验采用的评价指标为准确率,精确率,召回率,F1score以及AUC。
本实施例将本申请提供的多中心小样本场景下目标预测模型构建方法与现有经典的传统机器学习算法进行对比,对比算法包括:线性判别分析(Linear Discriminant Analysis,LDA)、k近邻(K-Nearest Neighbor,KNN)、朴素贝叶斯(Bayes,NB)、分类回归树算法(classification and regression tree, CART)、支持向量机(support vector machines,SVM)以及TSK模糊系统(Takagi-Sugeno-Kang Fuzzy-System,TSK-FS)的一阶形式和零阶形式。下表记录了这些算法十次实验五项分类评价指标的均值和方差(为了便于观察结果,所有结果均保留4位小数):
其中,AUC(Area Under Curve)被定义为ROC曲线下的面积。通常 使用AUC值作为模型的评价标准是因为很多时候ROC曲线并不能清晰的说明哪个分类器的效果更好,而作为一个数值,对应AUC更大的分类器效果更好。从表中可以看出,在准确率、精确率和AUC这三个指标上,本申请提供的多中心小样本场景下目标预测模型构建方法都比其他算法有着更好的表现,在召回率和F1分数上与其他传统算法也是可比较的,因此,从分类性能上看,本申请提供的多中心小样本场景下目标预测模型构建方法有着比其他传统算法更好的表现。
另外,考虑到算法的可解释性,实验中1-TSK-FS、0-TSK-FS以及本申请提供的多中心小样本场景下目标预测模型构建方法这三个基于模糊系统的规则复杂度比较如下表所示:
由上表可知,虽然0-TSK-FS比本申请提供的多中心小样本场景下目标预测模型构建方法有着更低的规则复杂度,但本申请提供的多中心小样本场景下目标预测模型构建方法的分类性能明显远远好于0-TSK-FS。而1-TSK-FS从分类性能上和规则复杂度上都不如本申请提供的多中心小样本场景下目标预测模型构建方法优秀。综上所述,本申请提供的多中心小样本场景下目标预测模型构建方法的分类性能和可解释性比这些传统算法都更好。
在模型正式训练时,通过观察每次训练结果不断调整交叉验证的参数寻优范围,避免过复杂的参数模型导致过拟合的发生,提高模型泛化能力。在进行多次训练后,逐步确立一个稳定、性能良好的模型。在实际应用中,只要有一部分机构有着充足的样本数据,对于样本量较少的机构也可以在不侵犯其他机构数据隐私的情况下快速,有效地进行本地部署,获得具有较好分 类能力、泛化能力以及可解释性的本地模型投入使用,协助多机构更加高效地完成联合训练。
显然,上述实施例仅仅是为清楚地说明所作的举例,并非对实施方式的限定。对于所属领域的普通技术人员来说,在上述说明的基础上还可以做出其它不同形式变化或变动。这里无需也无法对所有的实施方式予以穷举。而由此所引申出的显而易见的变化或变动仍处于本发明创造的保护范围之中。

Claims (8)

  1. 一种多中心小样本场景下目标预测模型构建方法,其特征在于,包括以下步骤:
    S1:在具有m个数据中心的多中心训练数据集中输入训练样本的特征,使用多个子分类器以及历史知识分类器分别对训练样本进行预测,所述历史知识分类器为融合了所有子分类器规则的分类器;整合所有预测结果共有m+1维,每个维度代表一个分类器的预测结果,大于0则判断为正类,小于0则为负类;
    S2:将预测结果向量归一化到[-1,1]之间;
    S3:以归一化后的预测向量为预测特征,结合训练样本对应的训练标签,使用岭回归方法训练每个子分类器的权重;
    S4:结合预测向量和训练得到的分类器权重得到综合预测结果;
    S5:对综合预测结果进行分类,得到与所述训练标签相同的结果;
    其中,第一子分类器的训练包括基础阶段和学习阶段;其他子分类器的训练基于第一子分类器的训练,包括学习阶段;所述基础阶段包括:借助经典0阶TSK来获取初始模糊规则参数c、Pg、δ以及错误分类样本信息errors,使用高斯核函数作为隶属度函数;其中,c和δ为前件参数,c是输入数据Data借助FCM聚类得到的聚类中心,δ是根据输入数据Data、FCM聚类得到的Data对聚类中心的隶属度矩阵U以及c得到的隶属度函数核宽度。
  2. 根据权利要求1所述的多中心小样本场景下目标预测模型构建方法,其特征在于,采用岭回归方法进行权重训练的目标函数以及权重参数的求解公式为:

    Pg=(λI+PTP)-1PTY,
    其中,P为预测向量,W为要训练的子分类器权重向量,Y为对应训练样本的真值标签,λ为结构风险项的正则化参数,Pg为通过岭回归方法计算得到的后件参数,I为单位矩阵。
  3. 根据权利要求1所述的多中心小样本场景下目标预测模型构建方法,其特征在于,所述学习阶段包括:先对c和Pg执行知识丢弃处理,然后进行知识整理得到最佳知识组合,最后结合输入数据Data计算出当前模糊规则对应的后件知识,得到预测结果。
  4. 根据权利要求3所述的多中心小样本场景下目标预测模型构建方法,其特征在于,所述知识丢弃处理包括:首先,确定知识丢弃比例r1和知识混乱特征比例r2,根据r1随机选择出要进行知识丢弃的规则,根据r2随机选择要进行知识混乱处理的特征列lose_columns;对于每一条要进行知识丢弃的规则,随机确定要进行知识遗忘或知识混乱;要进行知识遗忘的规则直接将c和Pg置空;对于要进行知识混乱处理的规则,将该规则的lose_columns列以其他规则的相对应列值进行替换。
  5. 根据权利要求4所述的多中心小样本场景下目标预测模型构建方法,其特征在于,所述知识整理包括:将丢弃处理后的c和Pg以及错误分类样本信息errors一起作为候选知识,从中选出最佳知识组合作为当前阶段的预备知识,所述错误分类样本信息errors包括分别对应c和Pg的样本特征和样本标签。
  6. 根据权利要求5所述的多中心小样本场景下目标预测模型构建方法,其特征在于,结合输入数据Data计算出当前模糊规则对应的后件知识,得到预测结果具体包括:结合输入数据Data和预备知识中的c计算出隶属度函数核宽度δ;利用前件参数c和δ以及对应后件参数的历史迁移知识,通过岭回归的方法,将经验风险项、结构风险项以及历史后件参数迁移知识正则化项作 为目标函数计算出当前模糊规则对应的后件知识,目标函数表达式以及后件参数求解公式如下:

    其中,Xg是高斯核映射后训练样本的输入向量,Pg-history历史后件参数迁移知识。
  7. 根据权利要求1所述的多中心小样本场景下目标预测模型构建方法,其特征在于,每个子分类器训练之后都将产生的模糊规则更新加入到历史模糊规则库中,以便后续子分类器的继续使用。
  8. 一种多中心小样本场景下目标预测方法,其特征在于,步骤包括:
    S1:采用如权利要求1-7任意一项所述的多中心小样本场景下目标预测模型构建方法构建预测模型;
    S2:在所述预测模型中输入当前目标中与训练样本特征对应的特征;
    S3:所述预测模型根据所述特征得出该当前目标的预测结果。
PCT/CN2023/116931 2023-07-04 2023-09-05 多中心小样本场景下目标预测模型构建方法和预测方法 Ceased WO2025007412A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/412,519 US12260350B2 (en) 2023-07-04 2024-01-13 Method for constructing target prediction model in multicenter small sample scenario and prediction method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202310807852.0A CN116596161B (zh) 2023-07-04 2023-07-04 多中心小样本场景下目标预测模型构建方法和预测方法
CN202310807852.0 2023-07-04

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/412,519 Continuation US12260350B2 (en) 2023-07-04 2024-01-13 Method for constructing target prediction model in multicenter small sample scenario and prediction method

Publications (1)

Publication Number Publication Date
WO2025007412A1 true WO2025007412A1 (zh) 2025-01-09

Family

ID=87604662

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/116931 Ceased WO2025007412A1 (zh) 2023-07-04 2023-09-05 多中心小样本场景下目标预测模型构建方法和预测方法

Country Status (3)

Country Link
US (1) US12260350B2 (zh)
CN (1) CN116596161B (zh)
WO (1) WO2025007412A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120031264A (zh) * 2025-04-22 2025-05-23 长春工业大学 一种基于小样本数据的电量预测方法
CN121365753A (zh) * 2025-12-23 2026-01-20 中电科大数据研究院有限公司 算力异构场景下实现分布式多模态大小模型协同训练方法、装置及系统

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116596161B (zh) 2023-07-04 2023-10-13 江南大学 多中心小样本场景下目标预测模型构建方法和预测方法
CN118536010B (zh) * 2024-07-12 2024-10-15 浙江大华技术股份有限公司 基于场景估计的感知数据处理方法、设备及存储介质
CN119479830B (zh) * 2025-01-15 2025-05-16 北京大学 一种多中心数据聚合方法
CN120448968B (zh) * 2025-04-27 2025-12-02 广东工业大学 一种基于多尺度特征学习与域自适应优化的小样本故障诊断方法
CN120069100B (zh) * 2025-04-29 2025-07-25 江南大学 一种分布式多中心的机器学习智能决策方法
CN120180097B (zh) * 2025-05-16 2025-09-02 内蒙古自治区农牧业科学院 草麻黄根单宁纯化用大孔树脂的筛选方法
CN120804753B (zh) * 2025-07-16 2026-02-24 山东德源电力科技股份有限公司 一种电能表实时故障智能检测方法
CN121190124B (zh) * 2025-11-21 2026-03-24 北京森博明德人工智能科技股份有限公司 Ai智能营销内容发布主体投放效果预测方法
CN121766730A (zh) * 2026-03-05 2026-03-31 北京亚洲卫星通信技术有限公司 基于历史工况自学习的铁路智能装车参数优化方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298382A (zh) * 2019-05-27 2019-10-01 湖州师范学院 一种基于ifcm、knn和数据字典的集成tsk模糊分类器
CN114298082A (zh) * 2021-11-09 2022-04-08 湖州师范学院 一种基于增强深度tsk模糊分类器的癫痫脑电识别方法
CN115186798A (zh) * 2022-06-09 2022-10-14 湖州师范学院 基于知识蒸馏的再生tsk模糊分类器
CN115186731A (zh) * 2022-06-09 2022-10-14 湖州师范学院 基于多层次特征融合的深度tsk模糊分类器
JP2023011313A (ja) * 2021-07-12 2023-01-24 株式会社Jvcケンウッド 推論装置、推論方法、および推論プログラム
CN116596161A (zh) * 2023-07-04 2023-08-15 江南大学 多中心小样本场景下目标预测模型构建方法和预测方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6757584B2 (en) * 2000-07-19 2004-06-29 Prudsys Ag Device and method for generating a classifier for automatically sorting objects
CN108846512B (zh) * 2018-06-05 2022-02-25 南京邮电大学 基于择优分类的水质预测方法
CN111738323B (zh) * 2020-06-12 2024-04-16 上海应用技术大学 基于灰色马尔科夫模型的混合增强智能轨迹预测方法和装置
CN112949895A (zh) * 2020-12-21 2021-06-11 华能新能源股份有限公司 一种基于动态可扩展神经网络模型的风速预测方法
CN113988418B (zh) * 2021-10-28 2026-02-06 上海师范大学 一种能源负荷预测的可视化方法
CN114067173B (zh) * 2021-11-07 2025-01-10 西北工业大学 基于多清晰度集成自训练的小样本低质量图像目标检测方法
US12211263B2 (en) * 2021-11-24 2025-01-28 Satsure Analytics India Private Limited System and method for assessing pixels of satellite images of agriculture land parcel using AI
CN114490985B (zh) * 2022-01-25 2023-01-31 北京百度网讯科技有限公司 对话生成方法、装置、电子设备和存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298382A (zh) * 2019-05-27 2019-10-01 湖州师范学院 一种基于ifcm、knn和数据字典的集成tsk模糊分类器
JP2023011313A (ja) * 2021-07-12 2023-01-24 株式会社Jvcケンウッド 推論装置、推論方法、および推論プログラム
CN114298082A (zh) * 2021-11-09 2022-04-08 湖州师范学院 一种基于增强深度tsk模糊分类器的癫痫脑电识别方法
CN115186798A (zh) * 2022-06-09 2022-10-14 湖州师范学院 基于知识蒸馏的再生tsk模糊分类器
CN115186731A (zh) * 2022-06-09 2022-10-14 湖州师范学院 基于多层次特征融合的深度tsk模糊分类器
CN116596161A (zh) * 2023-07-04 2023-08-15 江南大学 多中心小样本场景下目标预测模型构建方法和预测方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120031264A (zh) * 2025-04-22 2025-05-23 长春工业大学 一种基于小样本数据的电量预测方法
CN121365753A (zh) * 2025-12-23 2026-01-20 中电科大数据研究院有限公司 算力异构场景下实现分布式多模态大小模型协同训练方法、装置及系统

Also Published As

Publication number Publication date
US20250013898A1 (en) 2025-01-09
CN116596161B (zh) 2023-10-13
CN116596161A (zh) 2023-08-15
US12260350B2 (en) 2025-03-25

Similar Documents

Publication Publication Date Title
CN116596161B (zh) 多中心小样本场景下目标预测模型构建方法和预测方法
Biswas et al. An XAI based autism detection: the context behind the detection
Mottaqi et al. Contribution of machine learning approaches in response to SARS-CoV-2 infection
Comito et al. AI-driven clinical decision support: enhancing disease diagnosis exploiting patients similarity
Chadaga et al. Predicting cervical cancer biopsy results using demographic and epidemiological parameters: A custom stacked ensemble machine learning approach
Geetha et al. Evaluation based approaches for liver disease prediction using machine learning algorithms
Hegde et al. Symmetrized feature selection with stacked generalization based machine learning algorithm for the early diagnosis of chronic diseases
Chandrasekar et al. Lung cancer disease detection using service-oriented architectures and multivariate boosting classifier
Yothapakdee et al. Improving the efficiency of machine learning models for predicting blood glucose levels and diabetes risk
Hema et al. Disease prediction using symptoms based on machine learning algorithms
Agarwal et al. A mathematical model based on modified ID3 algorithm for healthcare diagnostics model
Singh et al. Real-time symptomatic disease predictor using multi-layer perceptron
Peya et al. Thyroid disease prediction based on feature selection and machine learning
Al-Batah et al. Improving oral cancer outcomes through machine learning and dimensionality reduction
Tusher et al. Semi-supervised learning: Assisted cardiovascular disease forecasting using self-learning approaches
Demirbilek et al. Recurrent multigraph integrator network for predicting the evolution of population-driven brain connectivity templates
Kommina et al. Post-stroke readmission prediction model using machine learning algorithms
Kumar et al. Hybrid optimization algorithms for accurate prediction and classification of cardiovascular diseases
Ainon et al. AMI Screening using linguistic fuzzy rules
Uddin et al. Review on diabetes detection using machine learning approaches
Xi et al. Transductive learning for bi-rads knowledge graph based on knowledge tensor factorization
Fawareh et al. Developing AI and ML models for real-world applications: Coronary artery disease
Mittal Healthcare data analysis using data mining techniques for disease prediction
Bindu et al. Comparison of novel machine learning algorithms for predicting chronic renal disease
Singla et al. Smart healthcare system for disease prediction using hybrid machine learning

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23944137

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 23944137

Country of ref document: EP

Kind code of ref document: A1