基于均衡系数的决策树优化算法

DECISION TREE OPTIMISATION ALGORITHM BASED ON EQUILIBRIUM COEFFICIENT

针对ID3算法多值偏向及误分类代价被忽视的问题,结合属性相似度和代价敏感学习,提出基于均衡系数的决策树优化算法。该算法既克服了多值偏向,又考虑了误分类代价问题。首先引进属性相似度和性价比值两者的均衡系数,对ID3算法进行改进;然后运用麦克劳林公式对ID3算法进行公式简化;最后将算法改进和公式简化相结合,得到基于均衡系数的决策树优化算法。实验结果表明,基于均衡系数的决策树优化算法,既能够提高分类精度,缩短决策树生成时间,又能考虑代价问题并降低误分类代价,还能克服多值偏向问题。

For the problems of ID3 algorithm in neglecting multi-value bias and misclassification cost,by combining the attribute similarity and cost sensitive learning we proposed an equilibrium coefficient-based decision tree optimisation algorithm. The algorithm solves the problem of multi-value bias and takes in to account the misclassification cost simultaneously. First,we introduced the equilibrium coefficient between attribute similarity and the value of performance to price ratio to improve ID3 algorithm. Then we used Mc Laughlin formula to carry out formula simplification on ID3 algorithm. Finally,we got the equilibrium coefficient-based decision tree optimisation algorithm by combining the algorithm improvement and formula simplification. Experimental results showed that the decision tree optimisation algorithm based on equilibrium coefficient can improve the classification accuracy,shorten the time of decision tree generation,and consider the cost problem as well as reduce misclassification cost. Moreover,it can also overcome the problem of multi-value bias.