Energy Valley Optimization with Very Fast Decision Tree for Credit Card Fraud Detection

In recent days, the Credit Card Fraud Detection (CCFD) for financial institutions has increased volume of transactions and evolving fraud schemes. The traditional models majorly struggled with handling complex data and lack of drift patterns in real-time transactions. Therefore, Energy Valley Optimization with Very Fast Decision Tree (EVO-VFDT) is proposed in this research. The VFDT model effectively handles the drift patterns in transactions and complex data with ability to adapt streaming data. Initially, the input features are collected from CCFD2023 dataset and undergo data exploration to construct a correlation matrix. Consequently, the correlation matrix is visualized as heatmap to find correlations between features. Then, the EVO metaheuristic algorithm is introduced to select optimal features by identifying most relevant features. Finally, the final subset of features is utilized to train the VFDT model to process the sequential transactions and adapts to real-time evolving data patterns. From the results, the proposed EVO-VFDT model attained better results when compared to Ensemble Soft Voting (ESV) model in terms of accuracy (99.80%), precision (99.65%), recall (99.92%) and F1-score (99.80%) respectively.