Performance Evaluation of Machine Learning-Based Algorithms to Predict the Early Childhood Development Among Under Five Children in Bangladesh

Abstract

In this research, an effort has been made to apply a number of classifiers to predict Early Child Development (ECD) in the context of Bangladesh using the Bangladesh multiple indicator cluster survey, 2019 data set (i.e., to evaluate which sort of algorithm best identifies ECDI). To predict the ECD, nine well-known machine learning algorithms were applied, including Linear Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Naïve Bayes (NB), Least Absolute Shrinkage and Selection Operation (LASSO), Classification Trees (CT), AdaBoost and Neural Network (NN). Children aged 48-59 months who were female, attending early education, reading three or more children's books, having playthings, having normal nutritional status, and were not disabled had a higher percentage of completing at least three childhood development domains, according to the bivariate analysis results. We found several performance parameters for the classification of early childhood development, including the following: Accuracy (LR) = 67.87%, AUC (LR) = 67.49%; Accuracy (RF) = 67.23%, AUC (RF) = 67.19%; Accuracy (SVM) = 67.37%, AUC (SVM) = 67.64%; Accuracy (NB) = 67.55%, AUC (NB) = 66.80%; Accuracy (LASSO) = 68.04%, AUC (LASSO) = 67.75. Based on the results of this investigation, LASSO regression predicts the ECD in Bangladeshi children moderately better than any other machine learning method utilized in this study.