Extraction of Bank Transaction Data and Classification using Naive Bayes

With the increase in the number of credit card transactions, particularly over the last few years, it is important to maintain a record of the corresponding Merchant Category Codes (MCCs) of these transactions. The benefits of doing so include being able to determine interchange fee, to determine payment types for tax purposes and so on. Data mining is used to process and extract useful information such as anomalies, patterns and relationships from a large bulk of data, including large transactional data. Classification can be used to analyse such data based on their MCCs and consequently use this information for a variety of applications. This processing of data can be made efficient by transforming the data to a suitable form for analysis using pre-processing measures. In this paper, an approach is presented to extract transactional data, pre-process using pattern matching and apply a Naive Bayes classifier to perform classification based on the MCC classes of the transactions. Evaluation of the model revealed an accuracy of 0.908 and error rate of 0.092 without any majority class assumption. With a majority class assumption, the model showed a precision of 0.927, recall of 0.883 and F-Measure of 0.904. These performance measures are very good, and indicates that the consideration of Naive Bayes as classifier was an optimal choice.