Unveiling Market Anomalies: Harnessing Convolutional Neural Networks for Fraud Detection in Finance

The process of financial fraud detection is essential for the sound functioning of financial markets. Conventional techniques cannot detect more complex forms of fraud, as such advanced techniques are necessary. This research aims to explore a new technique based on CNNs with attention mechanisms for improved fraud-detecting techniques. Transaction data from Kaggle which contain a diverse range of purchases and market trends were gathered and scaled through Min-Max normalization. The attention mechanism was used for feature selection and control over noise at the same time, focusing on important features. The developed CNN model has trained to identify the higher order feature of data and it shows a prominent performance of detecting the fraudulent activities. The proposed model has high accuracy of 99.12% which surpass traditional techniques and basic models like TransUNet, MedT, and FAT-Net by 19.36%. Systematic performance appraisal with proven examples as to how it may be applied served as the basis for this study. Based on the results of the current paper, the efficiency of CNNs in detecting market irregularities, and the importance of deep learning methods in fraud detection in finance are highlighted. This research was developed in Python to strengthen the security of financial systems, and ultimately promoting more trust and accuracy to the markets of finance across the globe.