Classification of smartphone users as adult or child in both constrained and non-constrained environments using mRMR-based feature selection and an ensemble classifier

A Smartphone is an important electronic device used by people of all ages. Excessive usage of smartphones among children can lead to various mental and physical problems. Hence, we believe that a control mechanism, if introduced, can help provide suitable content to users based on their age group. Our work focuses on detecting the age of the user based on their smartphone usage habits. To accomplish this, most of the previous work has collected datasets from users either in constrained or non-constrained environments. But in our work, we have collected data from both environments, and we were able to identify a generalized model to handle both environments’ data. To fill this research gap, we have collected our dataset while performing tasks such as typing, swiping, tapping, zooming, and measuring finger size. In a constrained environment, users must hold the phone either in their hands or on a table to finish the tasks. Whereas in a non-constrained environment, users are permitted to move freely while performing tasks. To achieve superior performance on both constrained and non-constrained data, we extracted some new statistical features, followed by Minimum Redundancy Maximum Relevance (mRMR) feature selection to select an appropriate set of features; the optimal feature count was identified using the cross-validation methods. We have used an ensemble classifier for classification, which takes a vote on the predictions of XGBoost, Random Forest (RF), and support vector machine (SVM). In our work, we have achieved 98.66% accuracy in constrained environments and 91.93% in non-constrained environments.