The fixed-size non-overlapping sliding window (FNSW) and fixed-size overlapping sliding window (FOSW) approaches are the most commonly used data-segmentation techniques in machine learning-based fall detection using accelerometer sensors. However, these techniques do not segment by fall stages (pre-impact, impact, and post-impact) and thus useful information is lost, which may reduce the detection rate of the classifier. Aligning the segment with the fall stage is difficult, as the segment size varies. We propose an event-triggered machine learning (EvenT-ML) approach that aligns each fall stage so that the characteristic features of the fall stages are more easily recognized. To evaluate our approach, two publicly accessible datasets were used. Classification and regression tree (CART), k-nearest neighbor (k-NN), logistic regression (LR), and the support vector machine (SVM) were used to train the classifiers. EvenT-ML gives classifier F-scores of 98% for a chest-worn sensor and 92% for a waist-worn sensor, and significantly reduces the computational cost compared with the FNSW- and FOSW-based approaches, with reductions of up to 8-fold and 78-fold, respectively. EvenT-ML achieves a significantly better F-score than existing fall detection approaches. These results indicate that aligning feature segments with fall stages significantly increases the detection rate and reduces the computational cost.

The current machine learning algorithms in fall detection, especially those that use a sliding window, have a high computational cost because they need to compute the features from almost all samples. This computation causes energy drain and means that the associated wearable devices require frequent recharging, making them less usable. This study proposes a cascade approach that reduces the computational cost of the fall detection classifier. To examine this approach, accelerometer data from 48 subjects performing a combination of falls and ordinary behaviour is used to train 3 types of classifier (J48 Decision Tree, Logistic Regression, and Multilayer Perceptron). The results show that the cascade approach significantly reduces the computational cost both for learning the classifier and executing it once learnt. Furthermore, the Multilayer Perceptron achieves the highest performance with precision of 93.5%, recall of 94.2%, and f-measure of 93.5%.

No abstract

No abstract