Predicting asthma control test score using machine learning regression models
Krishna Modi, Ishbir Singh, Yogesh Kumar · 2024
Asthma, a chronic respiratory condition, affects millions of individuals worldwide. Accurate prediction of Asthma Control Test (ACT) scores holds immense value for personalized healthcare interventions. This study embarks on the challenge of forecasting ACT scores by harnessing weather and demography features. Employing advanced feature engineering and rigorous data preprocessing, we investigate eight regression models: K-Nearest Neighbors, Logistic Regression, Decision Tree, Support Vector Machine (SVM), Random Forest, Deep Neural Network (DNN), ADA Boost, and XGBoost. Our primary objective is twofold: to evaluate the predictive ability of these models in the context of asthma and to conduct a comprehensive comparative analysis. The results unveiled each model&s;s efficacy and limitations in estimating ACT scores, paving the way for their real-world applicability. Among our findings, we observed noteworthy variations in model performance, with Random Forest and XGBoost emerging as the top-performing models due to their adeptness in minimizing prediction errors. Furthermore, our investigation illuminates the profound correlations between weather and demography features and ACT scores, underscoring the potential of environmental data in optimizing asthma management and control. This research offers valuable insights into the prediction of asthma control test scores based on weather and lifestyle features. Through comprehensive experimentation, we vary key hyperparameters such as the number of neighbours in KNN, the maximum tree depth in DT, the number of estimators in RF, kernel functions in SVM, and the number of iterations in DNN, among others. Our findings, coupled with the proposed method for future research, offer promising prospects for enhancing asthma management and personalized healthcare.