Regression clustering with lower error VIA EM algorithm

Abstract

Data mining is the process of analysing the any dataset to obtain understandable information. These information offer important details to improve decision processes about the data. Data mining is a technique that has a great potential to reveal significant information of consumer behaviours to companies. One of the most important tools for data mining is regression model. In this study, regression models of the dataset including the transaction numbers of several types of bank transactions within three year period are obtained. Besides, subclusters of the data set is generated by using expectation maximisation (EM) algorithm and regression models for every subclusters are created. Regression models of subclusters obtained via EM algorithm and regression model without clustering are investigated. By using root mean square error (RMSE) metric values comparison is made between these regression models. The results demonstrate that clustering the dataset by using EM algorithm creates regression models with lower error.