Machine learning prediction on number of patients due to conjunctivitis based on air pollutants: a preliminary study.

Abstract

A prediction of the number of patients with conjunctivitis plays an important role in providing adequate treatment at the hospital, but such accurate predictive model currently does not exist. The current study sought to use machine learning (ML) prediction based on past patient for conjunctivitis and several air pollutants. The optimal machine learning prediction model was selected to predict conjunctivitis-related number patients. The average daily air pollutants concentrations (CO, O3, NO2, SO2, PM10, PM2.5) and weather data (highest and lowest temperature) were collected. Data were randomly divided into training dataset and test dataset, and normalized mean square error (NMSE) was calculated by 10 fold cross validation, comparing between the ability of seven ML methods to predict the number of patients due to conjunctivitis (Lasso penalized linear model, Decision tree, Boosting regression, Bagging regression, Random forest, Support vector, and Neural network). According to the accuracy of impact prediction, the important air and weather factors that affect conjunctivitis were identified. A total of 84,977 cases to treat conjunctivitis were obtained from the ophthalmology center of the Affiliated Hospital of Hangzhou Normal University. For all patients together, the NMSE of the different methods were as follows: Lasso penalized linear regression: 0.755, Decision tree: 0.710, Boosting regression: 0.616, Bagging regression: 0.615, Random forest: 0.392, Support vectors: 0.688, and Neural network: 0.476. Further analyses, stratified by gender and age at diagnosis, supported Random forest as being superior to others ML methods. The main factors affecting conjunctivitis were: O3, NO2, SO2 and air temperature. Machine learning algorithm can predict the number of patients due to conjunctivitis, among which, the Random forest algorithm had the highest accuracy. Machine learning algorithm could provide accurate information for hospitals dealing with conjunctivitis caused by air factors.