Feature Selection Based Ensemble Support Vector Machine for Financial Fraud Detection in IoT

Although innovations like the Internet of Things, blockchain have simplified the process of making financial transactions, they have also given rise to new forms of fraud that have set back the advancement of Internet and IoT finance. The recent development of an ML-based financial fraud transaction system has highlighted the necessity for optimisation strategies to address the large dimensionality of the feature vector and the problem of class imbalance in any dataset. In this research, we offer a novel tactic to financial fraud detection that combines the bio-inspired optimisation algorithm with a two-stage ensemble support vector machine (ESVM). The dataset is first preprocessed through three stages that use the Bird Mating Optimisation Algorithm (BMOA) to ensure data balance. Finally, Least Square SVM (LS-SVM) and Latent Variable SVM (LV-SVM) are used for classification in the course of fraud detection. Furthermore, a comparison study between the suggested method and the current methods has been conducted. The suggested method effectively classified fraud transactions with an accuracy of 98%, far higher than the state-of-the-art approaches. The suggested method improves classification accuracy while simultaneously decreasing the number of incorrectly classified transactions and the expenses associated with misclassification. Compared to existing machine learning methods, the suggested BMOA-ESVM approach performs better in the tests.