Model for Prediction of Cataracts Using Supervised Machine Learning Algorithms

Abstract

This study identified the risk factors for cataracts and formulated a predictive model based on the identified variables. The study simulated the formulated model and validated the model with a view to developing a model for cataracts’ risk prediction. Following the review of the body of knowledge surrounding cataracts and their corresponding risk factors, interview with mental health professionals was conducted in order to validate the identified variables. Naïve Bayes, Decision Trees and the Multi-layer Perceptron classifiers were used to formulate the predictive model for the risk of cataracts based on the identified and validated variables using the WEKA software. The results of the data collected from 31 patients revealed 9 demographic variables and 17 risk factors variables alongside the respective risk factors, yielding a total of 26 variables in all. Out of the variables identified, the C4.5 Decision Trees algorithm revealed that smoking, myopia intensity, use of lenses and frequency of alcohol consumption were the most relevant risk factors out of cataract risks in the 26 variables identified. The results also showed that out of all the supervised machine learning algorithms used, the Multi-layer Perceptron was able to predict all records (100% accuracy) of the historical dataset used while the C4.5 Decision Trees and Naïve Bayes classifiers had an accuracy of 87% and 84% respectively. The study concluded that the Multi-layer Perceptron had the best capability to identify the unseen patterns existing within the variables used to formulate the predictive model for cataract risks.