Classification of Activities and Falls within a Multimodal Dataset

Accurately predicting fall detection from wearable sensor data has many implications. Detecting falls and other activities from wearable data provide a method by which it assists those who need it. Multiple methods were employed to predict activities from wearable data. One method was using a Recurrent Neural Network(RNN) known as a Long Short Term Memory network (LSTM). In addition, a traditional machine learning approach was explored with the use of a Random Forest Classifler(RFC). This work was adapted from a few previous works, as the dataset being used was that of the Challenge Up competition. Despite utilizing previous methods and works, the highest accuracy attained was 72% which lends itself to the potential difficulty of predicting rare events from time-series data.