AI-Based Digital Addiction & Screen-Time Behavior Analyzer

Abstract

This paper presents the design, implementation, and evaluation of an AI-powered system for real-time digital addiction detection and screen-time behavior analysis. The proposed system addresses the growing concern of excessive smartphone usage and its impact on mental health, academic performance, and workplace productivity. Existing solutions lack behavioral intelligence, real-time data extraction, and actionable intervention mechanisms. The platform integrates Android Debug Bridge (ADB)-based app usage extraction, a five-factor weighted machine learning (ML) risk scoring engine with sigmoid normalization, and a large language model (LLM)-powered insight generation module using Groq's LLaMA 3.3 70B to produce severity-rated behavioral recommendations. A Chrome Extension supplements the system by tracking browsing history in real time, enabling domain-level categorization and detection of inappropriate web content. Risk classification is performed across four levels — Minimal, Low, Moderate, and High — based on screen time, social media percentage, unlock frequency, night usage, and app concentration. A multi-channel alert system delivers warnings via ADB push notifications, ntfy real-time popups, and WhatsApp deep links when predefined thresholds are exceeded. The web dashboard renders hourly usage charts, app breakdowns, weekly trends, and a seven-day usage forecast powered by ARIMA time-series modeling. All data is persistently stored in Firebase Firestore with real-time synchronization. Experimental results confirm an average app usage detection accuracy of 97.5%, 100% risk classification accuracy across all four addiction levels, and sub-three-second AI response times. A dedicated ethics and privacy framework governs user consent, data storage, third-party API transmission, and compliance with GDPR and CCPA.