Researchers analyzed whether integrating wrist and hip accelerometer data with gyroscopic measurements could enhance the reliability of automated fall recognition systems deployed on smartphones and smartwatches. They found that Transformer-based neural networks combining these sensor inputs improve detection ability (F1-score rose 8% higher compared to using only wrist acceleration signals in offline tests). However, initial real-world trials resulted in many incorrect alarms. The team then recalibrated the system using labeled feedback captured during home-based monitoring. This adaptation cut erroneous alerts while maintaining high identification rates (reaching a 92% F1-score with consistent accuracy). The refined approach proved robust across different movement styles typical of older individuals