Paper Title
Machine Learning-Based on Device Contact Classification using Call and SMS Records

People enact different roles in society, for example, a woman can be a mother, sister, wife, aunt, friend, colleague, teacher, engineer, doctor etc. Unfortunately, many online social networking services reduce most people’s role to friends. In most devices, the contacts are stored by name and the relationship is not explicitly mentioned to save time and effort. A richer computational model capable of identifying relationships on the device would be useful for providing good recommendations, maintaining privacy, organizing communication and numerous other applications. This paper proposes a solution to strategically predict contact relationships by deploying an on-device model that uses available call, contact and SMS data to extract 126 features and make predictions on the device itself, avoiding any security issues. The machine learning model is able to classify relationships with 73% accuracy. The paper also presents a study of the different patterns of Call and SMS records during the Covid 19 lockdown and pre-Covid 19 lockdown and the effect of such societal factors in developing a machine learning model. By automating the contact classification procedure, this paper presents a unique solution in developing more personalized mobile systems in the future. Keywords - Machine Learning, Contact Classification, Mobile Communication, Call Records, SMS Records, On-Device.