Categorical Classification Approach for Identifying Multi-SIM Users from Call Detail Records

In this paper, we present a categorical classification approach for identifying multi-SIM users from Call Detail Records. Multi-SIM user classification is an unexplored domain in research literature and remains a challenging problem due to the diversity in telecom user population. This paper presents a subpopulation-based classification approach which incorporates this variety into the model, which is able to identify multi-SIM usage with higher precision and recall. A comparison of our approach to other baseline approaches (Gaussian Naive Bayes, Bernoulli Naive Bayes & Linear SVC) shows the effectiveness of subsample modelling for detecting multi-SIM usage. Additionally, we present an empirical study with which we quantify the contribution of oversampling and feature selection for multi-SIM detection. Further, using feature importance, we are able to identify possible rationales behind multi-SIM usage.