Supervised Machine Learning Model to Predict the Bank Loan Application Using Binary Classification, Decision Tree and Random Forest

This paper presents an online fraud detection system that uses anomaly detection to monitor an individual's behavior pattern and compare it with its usage history, which is a representation of the user's normal behavior patterns. Fraud is indicated by any significant deviation from normal behavior. The mechanism suffers from three disadvantages. Limited observations from the historical data, assorted nature of transaction data, and highly distorted information lead to unusually high positive failure rates of anomaly detection. Therefore, we propose a ranking metric embedding-based multi-contextual behavior profiling (ReMEMBeR) model to incorporate the detection mechanism effectively. We transform the original anomaly detection problem into a pseudo-recommender system problem and solve it using a ranking metric embedding-based method. With collaborative filtering, an individual could utilize information from similar individuals implicitly and automatically, which alleviates the individual's possible lack of historical data. By the ranking scheme, the model is trained to maximize the ability to distinguish between legitimate and fraudulent transactions. This helps to make full use of label information and, thus, solves the data skewness problem to the utmost extent. The proposed model integrates multi-contextual behaviour patterns, from purely local to more global ones. Evaluating transactions against multi- contextual behaviour patterns could reduce the error rate and, hence, could bring down the false positive rate. By creating a contrast vector for each transaction based on the customer's past behavior sequence, we profile the differentiation rate of each current transaction against the customer's behavior preference.