MTrust: Call Behavioral Trust Predictive Analytics Using Unsupervised Learning in Mobile Cloud Computing

Abstract

Filtering spam voice calls are a still major challenge in today’s technology contrary to SMS or email-based spamming. A numerical measure of the trust between users can help us filter calls based on relevance. Given the abundance of user-generated information available from the huge number of online devices, we can harness the power of this data to develop software adapting to user behavior. Existing research works for trust computation face various challenges when it comes to global applicability and understandability of trust values. Our investigation includes detailed surveillance of user call patterns based on the call data available from mobile devices and proposes a novel approach to filter calls that are of higher relevance to users based on their call-trust values. Our implementation realizes the diversity in call patterns of different people due to varying usage and uses classification and clustering algorithms to generate personalized, accurate numerical, and categorical trust values for every user. Categorical trust makes it easier to apply and understand trust ratings on a global scale. The implementation also incorporates a cloud facility to crowd-source trust values from multiple users, in a single database to generate the global trust of a user which can be used for spam filtering on a global scale. A software named “mTrust” is developed in this work for the future generation of a trustworthy mobile cloud network.