Real-Time Fraud Detection in Financial Transactions Using Support Vector Machines (SVM) and Data Mining Techniques

The real-time hybrid model of SVM and data mining, developed in this paper aims at detecting fraudulent financial transaction. All these have been developed in Python with Python Scikit-learn to adopt the SVM algorithm, the Pandas for data processing, and Seaborn and Matplotlib for deploying the results. Apache Spark and hence Spark SQL was used to ensure that the framework that handles the transactions data is highly scalable in real time. The data taken into account to this investigation are the IEEE-CIS Fraud Detection data set which includes 1,119,000 anonymized records of online credit card transactions. The class imbalance is a major problem in this dataset given that there are many more legitimate transactions than there are fraudulent ones. Therefore, in the present context, two stage concept was employed in the present study which include data preprocessing, feature extraction and integration of svm with clustering and anomaly detection for developing a system for recognition of not only the conventional and emerging frauds. The proposed hybrid model once again looks great with accuracy of 95.2 % of correct answers, a precise coefficient of 91.6 %, and recall coefficient of 90.7 % of true positive; this result is much higher from individual SVM or data mining models. Realtime performance analysis showed that average latency to be 100ms and throughput in terms of number of transaction per second as 800, which specifies that this model is effective for use in intensive traffic loads. Consequently, this work provides a feasible approach to conducting real-time fraud detection with potential for desirable scalability and practical recommendations for developing this aspect of FS security.