Detection of Roman Urdu fraud or spam SMS in Pakistan using machine learning

Abstract

over the past few years, mobile devices and their services have become widely used around the world. Almost everyone uses the Text Messaging Service (SMS) for communication purposes because it is easy to use and inexpensive. When a person tries to deceive another for the sake of profit (material or money), it is known as Fraud [1]. Through SMS fraud, fraudsters often adopt various strategies to make their messages look credible and legitimate. Various popular organizations use SMS services to advertise their products and send messages to individuals about their services. As a result, one receives many junk messages. Spam message is a message sent to any user who does not want to have it on their phone [2]. Spam or fraudulent messages can be threatening and can sometimes cause financial and confidential data loss. In Pakistan, messages are sent in English and Urdu (Pakistani national language) but most messages are sent using Roman Urdu (Urdu written using Latin / English characters). This research compares the strategies and algorithms used in the literature to detect spam / fraudulent messages written in English or in any local language such as Roman Urdu. The study also suggests a new way to detect fraudulent messages written directly in Roman Urdu. In the fraud detection process, three different monitoring machine learning classifiers are used in this study namely Support Vector Machine (SVM) [11], Naïve Bayes (NB) [12] and Decision Tree (J48) [13]. After using the model, we found that SVM performed better than the other two classifiers with 99.42% accuracy