MPredA: A Machine Learning Based Prediction System to Evaluate the Autism Level Improvement

Abstract

This paper describes the developmental process of a machine learning-based prediction system to evaluate autism Improvement level (MPredA), where the concerned user (parents or clinical professionals) can evaluate their children’s development through the web application. We have deployed our previous work (mCARE) data from Bangladesh for prediction models. This system can predict four major milestone parameter improvement levels of children with ASD. In this four-broad category, we have classified into four sub-milestones parameters for each of them to predict the detailed improvement level for each child with ASD. This MPredA can predict 16 milestone parameters for every child with ASD. We deployed four machine learning algorithms (Decision Tree, Logistic Regression, K-Nearest Neighbor, and Artificial Neural Network) for each parameter with 1876 data of the children with ASD to develop 64 prediction models. Among the 64 models, we selected the most accurate 16 models (based on the model’s accuracy and evaluation scores) to convert pickles file for the MPredA web-based application. For the prediction system, we have determined the most ten important demographic information of the children with ASD. Among the four-machine learning algorithms, the decision tree showed the most significant result to build the MPredA web-based application. We also test our MPredA -web application by white box testing and get 97.5% of accuracy with real data.