Improved C4.5 Rules Algorithm Based on Impact-Measurement of Suspected Instances
Xuegang Hu · Journal of Guangxi Normal University · 2007
Decision tree learning algorithm is a kind of classical classification algorithms in data mining field.Traditional decision tree learning algorithms treat all instances in the database in the same way,neglecting the differences of dubiety and impact among them,and the dirty data will distort the learning result,which will impact the quality of learning seriously.It provides in this paper an ameliorate C4.5 rules algorithm based on impact-measurement of suspected instances. Given a noisy dataset,first distinguishes the suspected instances from the dataset,then analyzes their impact and measures,last ranks rules for their covery of suspected instances,and classeifies the data based on the former steps.After re-ranking all rules,this algorithm minimizes the impact of dirty data and makes the knowledge of learning mostly closed to the correct data.The comparison between the traditional C4.5 rules algorithm and the algorithm presented here demonstrates the effectiveness of our strategies.