Description
Abstract—Medication non-adherence remains a persistent contributor to avoidable morbidity, hospitalizations, and
increased healthcare costs. This paper presents the design, implementation, and evaluation of an AI-enabled medication
adherence system centered on an IoT smart pill box that senses compartment openings and pill-removal proxies, detects
missed doses and adherence patterns, and issues escalating notifications to patients and clinical staff. The proposed system
combines (i) multi-modal sensing (magnetic lid/door switches, load-cell weight sensing, optional IR break-beam, optional
RFID/NFC identification, and time-stamped events), (ii) resilient connectivity (BLE to a mobile gateway with Wi-Fi/LTE
backhaul), and (iii) a hybrid inference pipeline that couples deterministic schedule rules with a learned time-series anomaly
detector for robust missed-dose detection under real-world variability (e.g., early/late dosing, compartment checks without
pill removal, and intermittent network outages). A pilot deployment (n = 32 participants, 28 days) demonstrates 93.8%
missed-dose detection accuracy with a 4.1% false-alert rate at a median end-to-end reminder latency of 18.6 s when the
gateway is present. The device sustained an average battery life of 31.4 days on a 2000 mAh cell under the evaluated duty
cycle. The results suggest that multi-sensor fusion and hybrid AI logic can improve reliability while preserving usability and
privacy in home monitoring scenarios.
Reviews
There are no reviews yet.