A New Weighted Associative Classifier Based on Nature Language Processing and Its Application in Chemical Data Mining

Associative classification mining (ACM) integrating association rule mining and classification has become a significant tool for knowledge discovery, especially in the chemical domain. Its major advantage is providing high accuracy as well as chemically interpretable models. Additionally, it is able to find associations among features while other traditional methods such as decision tree and naïve Bayesian consider the features independent to each other. In this paper, we propose a new weight framework for ACM based on information gain and graph theory. Combing this new scheme with CBA (classification based on associations), a novel classifier—IGAC (information gain and graph based associative classifier) is implemented and applied to three chemical datasets. In the generated models, the importance of the features related to the observed label classes is considered. The results show that not only IGAC can produce high accuracy (above 90%) but also the resulted models can be relatively easily interpreted by chemical knowledge. In addition, IGAC can discover meaningful rules which cannot be identified by classical associative classification mining (ACM).