一种采用渐进学习模式的SBS-CLearning分类算法

A classification learning algorithm of SBS-CLearning using step-by-step learning model

针对Learn++. NSE算法中多个基分类器之间相互独立、未利用前阶段学习结果辅助后续阶段学习而准确率较低的问题,借鉴人类的学习过程,优化Learn++. NSE算法内部的学习机制,转变基分类器的独立学习为渐进学习,提出了一种采用渐进学习模式的SBS-CLearning分类算法.分析了Learn++. NSE算法的不足.给出了SBS-CLearning算法的步骤,该算法在前阶段基分类器的基础之上先增量学习,再完成最终的加权集成.在测试数据集上对比分析了Learn++. NSE与SBSCLearning的分类准确率.试验结果表明:SBS-CLearning算法吸收了增量学习与集成学习的优势,相比Learn++. NSE提高了分类准确率.针对SEA人工数据集,SBS-CLearning,Learn++. NSE的平均分类准确率分别为0. 982,0. 976.针对旋转棋盘真实数据集,在Constant,Sinusoidal,Pulse环境下,SBS-CLearning的平均分类准确率分别为0. 624,0. 655,0. 662,而Learn++. NSE分别为0. 593,0. 633,0. 629.

The base-classifiers of the Learn++.NSE are separate,and the previous classifiers cannot help the forming of subsequent classifiers.The classification accuracy rate of the algorithm should be improved further.To solve the problem,drawing on the experience of the human learning process,the learning mechanism within the Learn++.NSE algorithm was optimized by the proposed SBS-CLearning gradual learning algorithm to transform the original independent learning of base-classifiers into a step-by-step learning.The disadvantages of Learn++.NSE were analyzed,and the process of SBS-CLearning algorithm was given.The incremental learning was conducted first on the basis of base-classifier,and the final ensemble result was then finished.The classification accuracy rates of SBS-CLearning and Learn++.NSE were compared based on the test data.The experimental results show that the SBS-CLearning has the advantages of both incremental learning and ensemble learning and can improve the classification accuracy compared with the Learn++.NSE.For the artificial SEA data,the average classification accuracy rates of SBS-CLearning and Learn++.NSE are 0.982 and 0.976,respectively.For the real rotating checkerboard data,under different Constant,Sinusoidal and Pulse environments,the average classification accuracy rates of SBS-CLearning are 0.624,0.655 and 0.662 with those of Learn++.NSE of 0.593,0.633 and 0.629,respectively.