Abstract:

The associative classification algorithm based on support and confidence is an important classification algorithm in data mining. This algorithm discovers frequent item sets and generates rules according to the threshold of confidence. However, the rules are of low quality. To address the problem, we propose an improved associative classification (AIAC) algorithm. Firstly, the AIAC selects a large number of attribute-value pairs to build small data sets. Secondly, the body of each rule is made up of the best attribute-value pairs picked from the small data sets. Finally, the AIAC employs the instance covering technique to cover all of the instances in small data sets, and builds a high quality classifier. Experimental results on 25 UCI datasets show that the AIAC can achieve much higher classification accuracy. 

Key words: data mining, associative classification, support, confidence, classification accuracy