Fraudulent Financial Transactions Detection Using Machine Learning

Abstract

there is a critical need to identify transactional risks in financial institutions for the purpose of improving customer experience and minimizing financial loss. This current study compares various machine learning (ML) algorithms with the aim of predicting legitimate financial transactions in an effective and efficient manner. Those algorithms include variety of machine learning models that covers all types of ML. The dataset was gathered from Kaggle website for that datasets. It consists of 6,362,620 records and 10 features. The top ML model was Random Forest Classifier using the unbalanced dataset attained an accuracy (99.97<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>), precession (99.96<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>), recall (99.97<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) and F1-score (99.96<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>). Meanwhile, the Bagging Classifier is the best classifier with balanced dataset with an accuracy (99.96<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>), F1-score (99.96<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>), precession (99.95<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) and recall (99.98<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>).