Window-size impact on detection rate of wearable-sensor-based fall detection using supervised machine learning

For a supervised machine-learning based fall-detection approach, data segmentation is needed to split the data sequence into several segments. Then features are extracted from each segment, where those features are used to train and test the classifier using supervised machine-learning algorithms. A sliding window is normally used to segment a data sequence. Although the sliding window takes an important role in the training and testing processes, most existing studies in fall detection rely on the figures from previous studies, without any studies to support them. This study investigates two types of the sliding window that are usually used by existing studies: Fixed-size Non-overlapping Sliding Window (FNSW) and Fixed-size Overlapping Sliding Window (FOSW). Two publicly-accessible datasets are used in this study. Two machine-learning algorithms, Support Vector Machine (SVM) and k-Nearest Neighbor (k-NN), are used to train and test the classifier. The experiments show that using a window size of 2 seconds for FNSW is recommended to achieve a better F-score, where using a 25% window overlap for FOSW is recommended to get a better precision. Overall, FNSW-based machine learning approaches can achieve up to 95.1% of F-score on average, while FOSW-based machine learning approaches can achieve up to 93.3% of F-score on average.