Classification of Emotions in Ambient Assisted Living Environment using Machine Learning Approaches

Abstract

By providing products and services that have an ambient environment, ambient assisted living (AAL) creates a secure environment for elderly people who are in need. The AAL helps a person’s health and well-being by systematically managing their daily routines and activities. AAL’s goals include allowing individuals to live independently in a preferred setting, keeping an eye on their health, and maintaining privacy and security. The individuals residing in the AAL environment exhibit various emotions while performing various activities. The traditional existing systems for emotion classification are based on facial and speech recognition techniques, which are useful in classifying the emotions of the people. From the literature, it can be found that only few research work used the machine learning techniques to classify the emotions of the people. Hence, this work aims at classifying the emotions using machine learning classification algorithms. The dataset used in this work is generated using the benchmark CASE (Sharma K, Castellini C, van den Broek EL, Albu-Schaeffer A, Schwenker F (2019) A dataset of continuous affect annotations and physiological signals for emotion analysis. J Sci Data 6(1):1–13. IEEE) dataset. The CASE dataset is generated by making 30 participants watch various videos and simultaneously capture real-time continuous annotation of emotions experienced using a novel, intuitive joystick-based annotation interface and also from eight sensors (respiration and skin temperature sensors, BVP, ECG, EMG (3x), GSR (or EDA)) that were attached to each participant. In this work, a new dataset is created by combining the above datasets. After preprocessing the dataset to scale the features and applying feature selection methods, various machine learning algorithms like KNN, SVM, logistic regression, random forest, etc. are applied on the dataset to classify four emotion classes – happy, sad, scared, and relaxed. This work shows that the proposed feature scaling and feature selection process improves the classification performance for the emotion recognition system. The experimental results show that random forest algorithm gives best results compared to other algorithms.