Potential of Automated Feature Engineering to Classify Credit Card Fraudulent Transactions of E-Commerce and Business in the Mobile Age
The primary objective of data analytics is to delineate hidden patterns and use them appropriately for taking strategic decisions in a variety of contexts in the organizations. Credit card frauds are intensifying significantly with the advancement and availability of state-of-the-art technology to everyone and become an easy target for fraudulent. Machine Learning algorithms and many Data Mining practices are already been used widely to detect fraudulent activities in credit card usage. But it is an open fact that credit card transactions are isolated events and cannot be considered as a sequence of transactions like normal transactions. In this article, the sequence of credit card transactions was considered in three different dimensions. Firstly, the sequence contains fraudulent and non-fraudulent activities, secondly, the sequence that is considered from fixed card-holders and fixed payment terminal, the third one is, the sequence of transactions that contains the amount incurred between the present and previous transactions. These three categories are considered in a combination to get eight unique binary perspectives of the transactions from the data set. These combinations are given as input to the Hidden Markov Model. HMM model results are again passed to the Random Forest classifier for further processing to investigate and detect fraudulent activities. The HMM model with multiple dimensions gives automated feature engineering to model temporal correlations and to enhance the effectiveness and accuracy of the classification process. In this article, it is demonstrated that the feature engineering strategy is well suited for e-commerce and a diverse set of businesses and also compared how the existing strategies deal with structural missing values.