Classification of type-2 diabetic patients by using Apriori and predictive Apriori

In this study a new approach to generate association rules on numeric data is proposed. It has been observed that equal binning techniques are not always useful to convert numerical data into categorical data, specifically in medical data. The proposed approach utilise a modified equal width binning interval technique to discretise continuous valued attributes to nominal based on opinion taken from medical experts. Approximate width of the desired intervals is chosen based on the advice given by medical experts and is given as an input to the model. Apriori algorithm usually used for the market basket analysis is used to generate rules on Pima Indian diabetes data. The study compares the quality of different association rule mining approaches for classification. The proposed approach utilises standard Apriori and predictive Apriori algorithms to generate association rules and highlights the importance of the often neglected pre-processing steps in data mining process. The proposed approach can help doctors to explore their data in a better way.