MITIGATING CREDIT CARD FRAUD IN EUROPEAN COUNTRIES

The detection of credit card fraud is essential to lower monetary losses and boost consumer trust in financial institutions. The effectiveness of two machine learning models, the Hidden Markov Model (HMM) and Logistic Regression, in identifying credit card fraud was examined in this study. Data was gathered from a large dataset of transaction records to analyze each model's projected accuracy, precision, and recall. The research questions addressed are: (Q1) In terms of recognizing credit card fraud, how do artificial neural networks and decision trees perform differently? (Q2) What are the comparative accuracy levels of Markov versus Logistic Regression when it comes to credit card fraud detection? These research questions were analyzed by using different machine learning models called ANN, Decision trees, Harkov Markov Model and Logistic regression. The findings and conclusions are (Q1) Both the Artificial Neural Network (ANN) and Decision Tree models achieved an accuracy of 99.97% in detecting fraudulent transactions. However, the Decision Tree model's performance is likely due to the dataset's quirks, suggesting that it might behave differently on datasets that are more complex or unbalanced. (Q2) The Logistic Regression model achieved a 99.83% accuracy rate in detecting fraudulent transactions. The Hidden Markov Model (HMM) showed a89% accuracy rate, effectively identifying fraud with a great precision score of 0.93 and a somewhat lower recall of 0.83. This indicates that while HMM may be useful for detecting sequential fraud patterns, it may not be as precise as logistic regression in detecting all fraud scenarios, especially when working with non-sequential datasets. Areas for further studies include: (a) concentrating on strengthening resilience on unbalanced datasets, (b) increasing model scalability, (c)decreasing computing complexity(d) enhancing fraud detection by investigating ensemble techniques, deep learning architecture (LSTM, RNNs), and hybrid models, and (e)further strengthening the accuracy and adaptability model by testing with more datasets, feature engineering, and real-time data.