Routine Predictor for Vimar VIEW: On-device behavioral pattern recognition in Smart Home systems

While Smart Home environments continue to evolve, the operational overhead of defining custom automations remains a significant barrier to an authentically seamless user experience. This thesis presents the design and implementation of a Routine Predictor engine integrated into the Vimar VIEW iOS application. The project transitions home management from a reactive, command-based model to a proactive environment that automatically learns user habits and control patterns through Edge Machine Learning (ML). The engine identifies recurring behavioral patterns directly on-device through a multi-stage pipeline combining statistical clustering and predictive scoring. To ensure the most suitable mobile integration, this work includes a comparative evaluation of alternative machine learning architectures, assessing classification performance against computational cost, latency, and resource efficiency on mobile hardware. This Edge ML philosophy ensures strict data privacy, minimal latency, and high reliability, as all sensitive behavioral data remains localized to the user's device while running only when the user actually interacts with the system. A key part of the system is the integration of a simplified Wizard and proactive touchpoints, such as Banners, which allow users to validate and finalize suggested routines with minimal friction. The final result is a system that learns the user’s habits without their data ever leaving the phone, and that asks nothing more than a confirmation before acting.