Measuring Engagement Level in Child-Robot Interaction Using Machine Learning Based Data Analysis

Abstract

In this paper, an approach for measuring the engagement level of a child interacting with a social robot in special education is proposed. The introduced methodology uses visual data gathered during the educational procedure and utilizes known machine learning models for analyzing the data. The examined models were able to estimate the engagement level by at most 14.70% Mean Absolute Error (MAE), with the Multi-Layer Perceptron (MLP) model showing the best performance with 12.70% MAE. Moreover, a Long Short-Term Memory (LSTM) model was able to predict the engagement level by a 10.40% MAE, when handling the problem as a time series prediction task. The overall results were very promising revealing the efficiency of the machine learning models to analyze the visual data describing the complex environment of the child-robot interaction in special education.