Fraud detection occurs a critical matter for businesses, financial institutions, and consumers, as fraudulent behaviour continues to increase in both complexity and frequency. Traditional methods of detecting fraud, while effective, often struggle with the scale and evolving nature of fraudulent behaviour. This study proposes a machine learning-based approach using decision trees to improve the accuracy. Decision trees, known for their interpretability and simplicity, are leveraged to classify operations as fraudulent or non-fraudulent constructed on a variety of input features. The model was trained and evaluated on a publicly available dataset, employing advanced preprocessing techniques such as feature selection and data balancing to handle class imbalances typical in fraud finding problems. The outcomes indicate that the decision tree model, when optimized, significantly improves detection rates while minimizing false positives likened to old rule-based techniques. This research demonstrates the potential of machine learning in combating fraud and offers insights into future improvements through ensemble methods and hybrid models.