1 Automated Human Activity Recognition by Colliding Bodies Optimization (CBO) -based Optimal Feature Selection with RNN Pankaj Khatiwada 1 Norwegian University of Science and Technology Teknologivegen 22, 2815 Gjøvik, Norway pankaj.khatiwada@ntnu.no Ayan Chatterjee 2 * University of Agder Jon Lilletuns Vei 9, 4879 Grimstad, Norway ayan.chatterjee@uia.no Matrika Subedi 2 University of Agder Jon Lilletuns Vei 9, 4879 Grimstad, Norway matris15@uia.no ABSTRACT In an intelligent healthcare system, Human Activity Recognition (HAR) is considered an efficient approach in pervasive computing from activity sensor readings. The Ambient Assisted Living (AAL) in the home or community helps people provide independent care and enhanced living quality. However, many AAL models are restricted to multiple factors that include computational cost and system complexity. Moreover, the HAR concept has more relevance because of its applications, such as content-based video search, sports play analysis, crowd behavior prediction systems, patient monitoring systems, and surveillance systems. This study implements the HAR system using a popular deep learning algorithm, namely Recurrent Neural Network (RNN) with the activity data collected from intelligent activity sensors over time. The activity data is available in the UC Irvine Machine Learning Repository (UCI). The proposed model involves three processes: (1) data collection, (b) optimal feature learning, and (c) activity recognition. The data gathered from the benchmark repository was initially subjected to optimal feature selection that helped to select the most significant features. The proposed optimal feature selection method is based on a new meta-heuristic algorithm called Colliding Bodies Optimization (CBO). An objective function derived from the recognition accuracy has been used for accomplishing the optimal feature selection. The proposed model on the concerned benchmark dataset outperformed the conventional models with enhanced performance. CCS Concepts • Information systems ➝ Information Retrieval • Computing methodologies ➝ Artificial Intelligence. Keywords Human Activity Recognition; Smart Activity Sensors; Optimal Feature Selection; Colliding Bodies Optimization; Recurrent Neural Network. 1. INTRODUCTION Human beings can perform numerous activities concurrently, such as walking, communicating, eating and many more. HAR identifies the goals and actions of humans from a series of observations on the actions of humans. The daily life activities of HAR include sitting, sleeping, walking, standing, etc. Various deep learning models, such as DBN, CNN, RNN, ANN, DNN, CRNN, etc., and machine learning algorithms such as SVM, NB, KNN, etc., are used in previous works. For identifying different levels of activities, such as walking, moving upstairs or downstairs, sitting, standing, lying, and running, machine learning and deep learning algorithms can be applied. The rapid growth of the Internet of things (IoT) has an advanced remote collection of human activity signals with the help of Wireless Sensor Network (WSN) [1][2]. The primary focus of our proposed activity recognition model is to detect the unexpected change in measures, such as covariance and mean that denotes the difference in an indoor environment [2]. The accurate manipulation of these measures is done using a robust algorithm. Activity recognition is relevant to the smart home model. For helping different emergency-related wellbeing services, such as fall detection for the elderly in healthcare, HAR is very helpful for aiding. Several physical activities are obtained for the nursing services and real-time responders in care homes and the domestic environment [1][3]. Introducing a scalable, robust, real-time indoor HAR model is a difficult task because of the complexity. The new technological developments and advancements stimulate the requirement for associated smart home sensing models and services [4]. In previous studies, different discrete classification models, such as Support Vector Machine (SVM), Naive Bayes (NB), K-Nearest Neighbor (KNN), Random Forest (RF), Conditional random fields (CRF), and Hidden Markov Model (HMM) have been used for human activity classifications, but deep learning algorithms can produce the better results as compared to the ML algorithms [5]. Some well-accepted deep learning algorithms used for HAR are Convolutional Neural Network (CNN), Long short-term memory (LSTM) and Recurrent Neural Network (RNN). To address the HAR and human pose recognition task, CNN is used in most applications by convoluting across two or three dimensions to seize a signal's spatial patterns [6]. HAR has become a lively and challenging research field in the past few years because of its applicability to various AAL domains to improve the demand for convenience services and home automation [7,12]. The camera usage is acceptable in most of the AAL scenarios. It provides various benefits in helping people with cognitive impairment, from event detection to person-environment interaction, assistive robots, and affective computing. The most typical applications of AAL are physiological monitoring, gait analysis, telerehabilitation, fall detection, and human behavior analysis. HAR has attained more interest in the ALL techniques in smart homes owing to the rapid increment of the world's aging population [7]. It is challenging to consider more count of observations in each second, the temporal nature of the observations, and the shortage of a straightforward procedure for relating accelerometer data for known movements [8]. Based on constant-size windows and