A Novel Approach for Continuous Authentication of Mobile Users Using Reduce Feature Elimination (RFE): A Machine Learning Approach

Abstract

Mobile phones are a valuable object in our daily life. With the acquisition of the latest technologies, their capabilities and demands increase day by day. However, acquiring the latest technologies makes mobile phones vulnerable to various security threats. Generally, people use passwords, pins, fingerprint locks, etc., to secure their mobile phones. Passwords and pins create so much burden for people always to remember their credentials. These traditional approaches are susceptible to brute force attacks, smudge attacks, and shoulder surfing attacks. Due to the difficulties mentioned above, researchers are leaning more towards continuous authentication. Therefore, this paper introduces an adaptive continuous authentication approach, a behavioral-based mobile authentication mechanism. In (Ehatisham-ul-Haq et al. J Netw Comput Appl 109:24-35, 2018), the authors achieved a good average accuracy of 97.95% with a Support vector machine classifier (SVM). We used LGB and RF and got 95.8% and 98.8% accuracy in user recognition. RF and LGB were trained for all five body positions separately to recognize each User among five users. This model also promises to reduce the system's cost and complexity by introducing the reduce feature elimination (RFE) technique during feature selection. RFE eliminates the less critical feature and reduces the dimension of the feature set. Hence, it demonstrates the benefits of our model for mobile authentication.