Classification On Baysican Fuzzy Learning And Fast Learning Based On Convex Hull Technique

Machine learning is an important component of artificial intelligence,which spans many disciplines,such as computer science,engineering technology and statistics.Classification techniques are important research topics in machine learning,as practical tools,which have been widely applied to face recognition,medical diagnosis,speech recognition document classification etc.Although the classification techniques in machine learning have received extensive attention,there are still many issues that need to be further explored and extensively studied,such as unbalanced data classification,improving the interpretability of clustering based fuzzy system and support vector machine based on convex hull technology for large-scale data classification.Therefore,we focus on the above issues in our study.The main contributions are as the following:1)When learning from imbalanced datasets,the tendency is that classical fuzzy classifiers might obtain a high predictive accuracy over the majority class,but might predict poorly over the minority class.In Section 2,a novel zero order Takagi-Sugeno-Kang fuzzy system is presented to improve the classification performance and rule-based interpretability for imbalanced datasets.In antecedent parameter learning stage,a new clustering method,called Bayesian fuzzy clustering based on competitive learning is proposed to partition the input space for the antecedents of if-then rules.In consequent parameter learning stage,based on the maximum separation strategy and by keeping the distance between the minority class and the classification hyperplane larger than the distance between the majority class and the hyperplane.Since considering the repulsed force of clustering prototypes between different classes and effectively correcting the skewness of the classification hyperplane,the proposed fuzzy system can obtain the satisfactory classification performance with high interpretability.Secondly,based on the study of Bayesian fuzzy clustering based on competitive learning,a cross-class Bayesian fuzzy clustering algorithm BF3 C is proposed in Section 3,BF3 C considers repulsion forces between cluster centers belonging to different classes to obtain more interpretable fuzzy space partition.BF3 C uses a particle filter algorithm to compute the optimal values of the number of fuzzy clustering,fuzzy membership and cluster centers for different classes.Then a novel TSK fuzzy classifier is presented to improve the classification performance and rule-based interpretability for imbalanced datasets.In order to improve the classification performance for imbalanced datasets,an imbalance learning algorithm is derived to estimate consequent parameters of fuzzy rules on the basis of the weighted average misclassification error.2)In order to design a TSK fuzzy classifier with satisfactory classification performance and high interpretability,a new TSK fuzzy classifier is proposed in Section 4.The proposed TSK fuzzy classifier uses a joint likelihood probabilistic model to characterize fuzzy classification under a Bayesian inference framework,and then simultaneously learn both the antecedent parameters and consequent parameters of fuzzy rules by applying the Markov-Chain Monte-Carlo technique.In terms of the maximum a priori principle,B-TSK-FC can guarantee the global optimality for the learnt parameter set.Since B-TSK-FC establishes an intrinsic link between the input and output space and reveals the structure of the whole dataset,it obtains the satisfactory classification performance and high interpretability(with less number of rules).3)Section 5 and Section 6 mainly discuss the classification problem of support vector machine combined with convex hull technique on large scale datasets.Firstly,in Section 5 facing the problem that current SVMs for large-scale datasets classification problems are almost sensitive to noises,a soft kernel convex hull support vector machine for large scale noisy datasets,called SCH-SVM,is proposed based on the soft kernel convex hull and pinball loss function.By finding the hyperplane of training samples of soft kernel hull in quantile distance,SCH-SVM obtains an anti noise classifier model.Moreover,four theorems prove that the soft convex kernel selection method of SCH-SVM is effective in classifying accuracy,the number of suport vectors and training time.In Section 6,a fast convex-hull vector machine(CHVM)is developed for large-scale ncRNA data classification tasks.By projecting a dataset onto all the corresponding two-dimensional projection combinations,CHVM extracts the convex hull vectors for the whole kernelized training set.Then,the convex hull vectors are presented as the inputs to a SVM classifier regardless of the adopted SVM's formulation.Since the scalability of CHVM is independent on SVM formulations,various types of SVMs can always be adopted in CHVM for the ncRNA data classification.