Two Unorthodox Aspects in Handcrafted-feature Extraction for Human Activity Recognition Datasets

Abstract

Human Activity Recognition (HAR) is one of the underlying research areas in the field of biomedical data science. One of the major tasks of researching with HAR datasets is extracting features from the raw datasets. Many methods have been proposed so far as an approach. In this paper, we have used two unorthodox methods (in the HAR field) for feature extracting, namely- ‘Vector Point’ (VP) and’ Absolute Distance’ (AD). By using these techniques, the data size gets decreased 3 and 6 times consecutively and the time complexity also gets decreased considerably. The extracted features are utilized as inputs to a Random Forest classifier. We have got an AVC score of ‘1’ for our merged dataset with both the VP and AD methods which manifests the success of our work.