Phone based fall detection by genetic programming

Elderly people are prone to fall due to the high rate of risk factors associated with ageing. Existing fall detection sys- tems are mostly designed for a constrained environment, where various assumptions are applied. To overcome these drawbacks, we opt to use mobile phones with standard built- in sensors. Fall detection is performed on motion data col- lected by sensors in the phone alone. We use Genetic Pro- gramming (GP) to learn a classi er directly from raw sensor data. We compare the performance of GP with the popu- lar approach of using threshold-based algorithm. The result shows that GP-evolved classi ers perform consistently well across di erent fall types and overall more reliable than the threshold-based.