Machine Learning Algorithm for Fall Classification Using Wearable Device
Md Wasif Islam Wasi, Rudzidatul Akmam Dziyauddin, Nur Izdihar Muhd Amir, Robiah Ahmad · 2022
Significant injuries that sometimes cause fatality due to fall, is even severe for elderly who lived alone. The goal of this paper is to investigate the use of supervised machine learning (ML) algorithms in classifying fall and normal states. Based on our previous 2245 dataset from ten subjects, we employed three types of ML namely, k-Nearest Neighbour (k-NN), support vector machine (SVM) and also decision tree (DT). Initially, the optimal$k$value of k-NN, 6, is determined by comparing a range of$\boldsymbol{k}$values in terms of mean error. The$\boldsymbol{k}$value equal to 6 is then used in the k-NN and a comparative study is conducted. Results show that the DT outperforms in sensitivity, 98% and also the accuracy, 98.52%. However, the k-NN outperforms in terms of specificity, 99.31 % when it detects the normal status. On the other hand, SVM demonstrated the lowest performance probably due to the data scaling issue.