Basman M. Hasan Alhafidh, et. al. International Journal of Engineering Research and Applications www.ijera.com ISSN: 2248-9622, Vol. 10, Issue 11, (Series-I) November 2020, pp. 31-37 www.ijera.com DOI: 10.9790/9622-1011013137 31 | Page Design and Implementation of Home Autonomous System Based on Machine Learning Algorithms Basman M. Hasan Alhafidh 1 , Amar I.Daood 2 , Modhar A. Hammoudy 3 1 Dept. of Comp. Engineering College of Engineering University Mosul, Iraq 2 Dept. of Comp. Engineering College of Engineering University Mosul, Iraq 3 Alhammoudy Dept of Comp. Engineering College of Engineering University Mosul, Iraq ABSTRACT Home automation systems are cutting edge tech- nologies to monitor and control a smart home environ- ment to produce an efficient system that accurately predicts the needs of the human occupants. Past research has focused on the accuracy of prediction of a users future action. However, a focus on prediction accuracy often comes at the cost of slower processing time. additionally, a need of hardware implementation is necessary to assure the consistency of the simulation results. Finally, much of that work uses synthetic datasets which do not always reflect the real-world interactions that occur between an individual and the home environment. This paper focuses on the prediction of future human actions in an intelligent environment with the goal of achieving both high prediction accuracy and response times that are appro- priate for a real-time application environment. Using several different machine learning algorithms, both the simulation experiments and hardware implementation were accomplished using the MavPad dataset which was gathered from a fully-instrumented home environ- ment. In the first stage of this study investigates which machine learning algorithm will satisfy the conditions of real-time application. The findings show that neural network technique provides a feasible solution in term of accuracy. Going deeper, the authors use simulation to investigate the performance of a multilayer neural network that predicts future human actions. In the second stage of this work the authors present a hardware implementation of the deep learning model on a FPGA. The results showed that the hardware implementation demonstrated similar accuracy with significantly improved performance compared to the software- based implementation due to the exploitation of parallel computing and using optimization tech- niques to map the designed system into the target device. Furethermore, our implementation of FPGA- based neural network system supports its future uti- lization for other applications. Keywords: Smart Home System, FPGA, Autonomous Sys- tem, Machine Learning Algorithms, Prediction System --------------------------------------------------------------------------------------------------------------------------------------- Date of Submission: 28-10-2020 Date of Acceptance: 09-11-2020 --------------------------------------------------------------------------------------------------------------------------------------- I. INTRODUCTION Consumers seek home automation systems that monitor and control a smart home environment to produce an efficient system that accurately predicts the needs of the human occupants. Some recent researches have presented the use of Machine Learning Algorithms (MLAs) to predict the present activity of an individual depending on sensors reading, and other works focused on the accuracy of prediction of next user action on home appliances. However, those studies used synthetic datasets that were generated based on assumptions which may not reflect the real-world interactions between an individual and their home environment. Also, there was often a focus on using MLAs that presented greater accuracy in prediction without considering the time constraints of real-time applications. In this paper, we will not focus on the creation of models of a user’s current activities or on generalizing occupant behavior inside the environment. Instead, we will focus on the prediction of next human actions on changes to the state of actuators inside an intelligent environment. Hence, the prediction process in such an environment needs a fast response speed from the designed prediction system with a high level of accuracy to satisfy the quality of the service based on real-time application domains. To assist the mentioned criteria, several simulation experiments were conducted to compare the average accuracy and average time of prediction results between different approaches of Nave Bayes (NB), Support Vector Machine (SVM), and Neural Network (NN). The implemented MLAs were applied to the MavPad dataset. This dataset was collected from sensors and actuators which distributed across a real- RESEARCH ARTICLE OPEN ACCESS