Effective Ways to Overcome Classification Limitations for Activities of Daily Livings (ADLs)

Abstract

Development of reliable and precise methods of Human Activity Recognition (HAR) are highly important, since wrong or inaccurate recognition can cause harmful consequences for human health. Scientists working in the field try to find the ways to enhance achievements for recognition accuracy. Taking this into account, it is vital to choose classifiers which make classification of activities with reliable rates. However, limitations of the currently existing algorithms and inherent lack of precision level can put their applicability to the field under the question. In particular, Hidden Markov Model has certain restrictions, which is caused by the principle of random selection of parameters and it is problematic to discriminate between the classes with high accuracy. Other well-known methods also have some drawbacks due to their nature. With the aim to solve these problems and to ensure the required results, we propose a hybrid complex of classifiers that provide adequate models of the study area. This work will deal with the classification of daily living human activities using wearable inertial sensors. Walking, Lying, Standing up, Stair Ascent, Stair Descent, Sitting on the Ground, etc. are examples of these activities. In this study, a dataset including twelve activities is created using three inertial sensors. Two hybrids of classifiers that combines Bayesian Networks with distance based classifier, namely with k Nearest Neighbor and Neural Network with Hidden Markov Model will be presented in the study. The achieved results will be compared in terms of correct classification rate, F-measure, recall, precision, and specificity. Raw data and extracted features will be used separately as inputs of the classifier. The inertial sensor units worn by different healthy subjects are placed at key points of upper/lower body limbs (chest, right thigh and left ankle). In this study, only acceleration data is used, as a modality for estimating the activities.