Detecting Fraud Calls vis-à-vis Natural Language Processing

Law defines fraud as the deliberate use of deception to seek unjust or unlawful advantage or to deprive a victim a right that they are entitled to. Fraud may be motivated by financial gainor other advantages, such as obtaining a specific document, identity takeover, etc. The ability to detect fraudulent behaviors on a large scale is made possible by the combination of massive data gathering and predictive or forensic analytics, which employ digitized data to reassemble or identify fraud. Scam calls, quite common today, are false calls that persons or businesses make to deceive recipients into parting with their money or private information. Scammers frequently play down the situation by calling it a normal call or even by lying to the person they were trying to contact. In this work, the objective is to detect fraud calls, employing Supervised Intelligent Learning Algorithms. For the same, a total of 5925 Data points are used to train the baseline, thereby arranging the model parameters to best fit the scenario. For obvious reasons, the source of the data is kept as anonymous. This work would help in the early detection of scam calls and would help a lot of innocent lives from getting tied by the knot of fraudulence. Different ML Algorithms have been used for the work, namely Random Decision Forest (RF), k - Nearest Neighbours (k-NN), Support Vector Machine (SVM), and Gaussian Naive Bayes (GNB). Further, these Algorithms have been compared based on their performance of prediction. According to the results observed in the research, k-Nearest Neighbour and the Support Vector Machine are best for the detection of Fraud Calls, as the model modeled using the k-NN, and the SVM Classifier gives an accuracy of 99%.