Affiliations: [a] Department of Computer Science, National University of Computer & Emerging Sciences, Islamabad, Pakistan. E-mails: [email protected], [email protected], [email protected] | [b] Department of Computer Science, Air University, Islamabad, Pakistan. E-mail: [email protected]
Abstract: Smartphones, with built-in array of sensors, provide an opportunity to ubiquitously collect user’s behavioral data. This leads to variety of founding applications that identifies interesting patterns in the smartphone data to learn human behavior. In this paper, we propose an approach that enhances the productivity of individual’s by unobtrusively learning their routine through smartphones. We design and develop a non-intrusive smartphone app – Prodmapp that periodically collects sensing data from user’s smartphone. We extract several potentially useful behavioral features from the data and perform correlation analysis among the features and user’s productivity score (ground truth). We collect 15 days sensing data from 10 users through Prodmapp. Ground truth is collected from the users in the form of questionnaires to quantify their productivity. The results showed that there exists a significant correlation among several behavioral features and user’s productivity score. Finally, we train and evaluate a prediction model using significantly correlated features that can predict the change in productivity of users by analyzing the variation in feature values. We train three classifiers i.e., logistic regression, SVM and KNN to compare their performance on the two benchmark datasets, one collected through Prodmapp and other from CASAS smart home project. Results shows that our proposed approach performs well and all three classifiers achieve good prediction accuracy on both datasets.
