Features Selection For Fall Detection Systems Based On Machine Learning And Accelerometer Signals

Fall among older people is a major medical concern. Fall Detection Systems (FDSs) have been actively investigated to solve this problem. In this sense, FDSs must effectively reduce both the rates of false alarms and unnoticed fall. In this work we carry out a systematic evaluation of the performance of one of the most widely used machine learning supervised algorithm (Support Vector Machine) when using different input features. To evaluate the impact of the feature selection, we use Area Under the Curve (AUC) of Receiver Operating Characteristic (ROC) Curve as the performance metric. The results showed that with four features it is possible to obtain acceptable values for the detection of falls using accelerometer signals obtained from the user's waist. In addition, we also investigate if the impact of selecting the features based on the analysis of a dataset different from the final application framework where the detector will be operative.