User’s Gender Prediction Based on Smartphone Applications Installed: Analysis from Real World Data to Simulation

Users’ gender information for a smartphone application cannot be approached directly since it is considered as private. However, this basic demographic information is of great value during persona analysis. In this paper, an innovative feature selection approach is proposed for gender prediction based on smartphone applications installed. We construct weighted average female-oriented and male-oriented features, and applying it to four machine learning algorithms. Meanwhile, simulation is carried out to make horizontal models comparison easy even between supervised and unsupervised algorithms. Prediction accuracy can be as high as 90.4% for weighted average logistic regression model.