Classification of Fraud Calls by Intent Analysis of Call Transcripts

While the rapid growth of technology makes life easier for consumers it has also brought new threats along with it. People can be duped to reveal sensitive data such as banking and credit card details, personally identifiable information, and passwords. One such technique is phishing through phone calls. The number of people who fall for such scams is an astounding figure. The conventional approach to detecting phishing call fraud depends on a blacklist of known fraud numbers. This creates a problem when new numbers or numbers which have not been encountered by the system before are used. To solve this, we propose a system that will classify a fraudulent call by analyzing the conversation between the potential victim and the caller. We used various machine learning techniques to perform intent analysis of call transcripts. We created two models and compared them. The models based on the Naive Bayes Algorithm and CNN have an accuracy of 94.57% and 97.21 % respectively.