The research work to predict the side effects of drugs is done by the relationship between chemical and target protein. Yamanishi et al. [7] proposed a kernel regression model as computational model. This method predicted a potential side effect profiles based on the chemical structures and the information on the target protein. This method ranked first one of the side effects of 41.7 % (275 drugs) in 658 drugs, and ranked a correct side effect among the top five scoring for 70.0 % (461 drugs) in 658 drugs. Kuhn et al. [8] focused on similarities in the side effects of drugs. The clinical data on chemical and protein were collected. And by combining drug-target protein and drug-side effect relations, overrepresented target protein-side effect pars are identified. 732 of 1428 side effects were predicted to be mainly caused by individual proteins. 137 of 732 side effects were proved by pharmacological or phenotypic existing data. From the results of these studies, classification and regression characteristics of the side effect are found. However, these studies will fall into a lack of an overall service-based solution because all data for the side effect prediction are not always prepared. Although the above-mentioned 2 studies focused on potential side effect, we focus not only on side effect, but also on its incidence. Therefore, we propose the comprehensive prediction model of drug side effect and its incidence using the data mining approach at the intersection of statics, machine learning and database system considering classification and regression characteristics of the side effect. And the gene information of target protein and the chemical (drug) - protein interaction are considered in this paper.