Subscriber Gender Prediction in Telecom Using Deep Learning

Organizations are changing their strategies to be called as data-driven. With this they get the benefit of estimating achievable targets, predicting budget allocations and goal-settings. But the true challenge lies in how accurate the considered data is. In this work, we propose various ways to derive one of the most sought after user information, the gender of a user from his/her presence in a telecommunications system. We considered the data from reality shows with prize money where gender is supposed to be mentioned correctly. We considered the user call pattern he/she does to find gender specific behaviour. We considered the app usage data like games and the type of Advertisements (Ads) a user reacts, to understand the uniqueness. We also considered browsing data of a user to highlight a significant difference in their behavior. Based on deep learning techniques, a model is built to apply the above sources as input to predict the gender with an accuracy of 81.2 percent. The major contribution in this work is towards predicting the gender from various telecom data sources of huge data sizes to a high accuracy.