Sustainable software engineering in practice : design and evaluation of an on-device rule-based recommendation system for digital carbon footprint awareness

Abstract

Digital services, particularly social media and content platforms, contribute significantly to global carbon emissions, yet most users remain unaware of the environmental impact of their daily usage. This thesis aims to address this gap by designing and evaluating an on-device recommendation system that makes digital carbon footprint visible and actionable for individual users. The work follows the Design Science Research methodology, applying sustainable software engineering principles throughout. The system was implemented as an Android application that passively records foreground usage time for ten social media and content platforms using the UsageStatsManager API and calculates CO₂ emissions using per-minute emission rates reported by (Iryna Komazova, 2024). A rule-based recommendation engine runs entirely on the device and generates up to three personalised suggestions: recommending shifts to lower-emission apps, detecting usage trends, nudging users when daily usage exceeds their average, and translating emissions into everyday equivalents such as car kilometres driven. The choice of rule-based reasoning over machine learning keeps computational overhead negligible and ensures full transparency. A four-week deployment with 37 participants evaluated the system in real-world conditions. The system ran reliably across all devices, and the recommendation engine correctly identified each participant's dominant high-emission application. The evaluation revealed substantial disproportionality between time share and CO₂ share: YouTube accounted for 24.5 percent of usage time but only 13.4 percent of emissions, while Reddit accounted for 5.3 percent of time but 15.8 percent of emissions, confirming that the system addresses a real awareness gap. The thesis contributes design knowledge for energy-aware on-device AI systems in the digital sustainability domain.