"Nowadays, sensory data have enriched (and also increased the complexity) of many real-world problems. Some of these problems aim to classify the current state with a label, others try to predict the next period values, while others try to identify failures or abnormal scenarios. The gathered Time Series (TS) can be either univariate or multivariate, according to how the signals are grouped by sub-systems.
One of these fields is the Fall Detection (FD) problem using wearable devices. The most common solution is to use a tri-axial accelerometer and some Machine Learning and Computational Intelligence to accurately detect the fall event. However, the problem comes from the fact that there is almost no data from real fall events; therefore, learning these models is in compromise.
This talk focuses on three different aspects of the research that have been studied to solve the FD problem with on-wrist wearables: the FD problem itself, the multivariate TS data sets balancing techniques and the multivariate TS clustering. Recent research findings will be explained for each of these topics, leading to the next future topic to study."