Identifying Best Machine Learning Model For Prediction Of Cataract Status Using Network Information Criterion

Abstract

Cataract is an eye condition that makes people very difficult to see things clearly. Previous studies reviewed on Machine Learning (ML) used cross-validation, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) to select best Machine Learning (ML) model for prediction of Cataract but the use of Network Information Criterion (NIC) remain a challenge. The data used was from on-going research by the corresponding author on eye diseases and it was collected through personal interview at National Eye Centre (NEC) Kaduna from 25^th Nov. to 19^th Dec. 2024 after ethical approval was given on 21^st October, 2024. The study computed the NIC values of Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbour (K-NN), Naïve Bayes (NB) and Multilayer Perceptron (MLP) ML models that are used to predict Cataract status in the training and test sets and it was discovered that DT was the best model to predict the disease because it has minimum NIC value of 0.0288. This study demonstrates the appropriateness of NIC in identifying the best model for prediction of Cataract status and this would help healthcare professionals in making accurate informed decisions. Future study should compare the values of AIC, BIC and NIC and select the best model for prediction of Cataract status for more reliable prediction of the disease.