A Minimalistic Approach to Predict and Understand the Relation of App Usage with Students' Academic Performances

2025-08-22

AI Summary

Prior studies rely on self-reported data, introducing subjectivity bias, and require longitudinal records, hindering early academic intervention. Method: This paper proposes a lightweight, real-time academic performance prediction framework leveraging authentic mobile application usage logs—specifically, session durations across app categories over the past seven days—to train machine learning models for predicting students’ cumulative grade point average (CGPA). Contribution/Results: Significance testing (p < 0.05) reveals strong positive correlations between CGPA and usage duration of productivity and book-related apps, and a significant negative correlation with video-app usage. High- and low-CGPA students exhibit markedly distinct app-category preferences and usage patterns. Evaluated on real-world data from 124 students, the model achieves a mean absolute error of 0.36 and an inference latency of only 0.31 seconds per prediction, demonstrating high accuracy, real-time capability, and deployability—establishing a novel paradigm for precise, early-stage academic intervention.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Cognitive Modeling & Cognitive Systems: Applications
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Due to usage of self-reported data which may contain biasness, the existing studies may not unveil the exact relation between academic grades and app categories such as Video. Additionally, the existing systems' requirement for data of prolonged period to predict grades may not facilitate early intervention to improve it. Thus, we presented an app that retrieves past 7 days' actual app usage data within a second (Mean=0.31s, SD=1.1s). Our analysis on 124 Bangladeshi students' real-time data demonstrates app usage sessions have a significant (p<0.05) negative association with CGPA. However, the Productivity and Books categories have a significant positive association whereas Video has a significant negative association. Moreover, the high and low CGPA holders have significantly different app usage behavior. Leveraging only the instantly accessed data, our machine learning model predicts CGPA within 0.36 of the actual CGPA. We discuss the design implications that can be potential for students to improve grades.

Problem

• Predicting academic performance using short-term app usage data
• Identifying app categories' impact on student grades
• Enabling early intervention with minimal data requirements

Innovation

• Uses 7 days actual app usage data
• Machine learning model predicts CGPA accurately
• Identifies specific app categories affecting performance