Machine Learning based Analysis and Prediction of Asthma Disease

Abstract

Asthma is considered to be one of the most common chronic pulmonary diseases across the world, which calls for appropriate identification and risk assessment to enhance patients' prognosis. Machine learning approaches have proved to be effective means of analyzing patients' data to predict a disease. This paper is aimed at comparing the effectiveness of four supervised machine learning approaches, such as logistic regression, decision tree, random forest, and K-Nearest Neighbor (KNN) for predicting and analyzing the disease of asthma. The experiment involves the analysis of a big Kaggle database with 316,800 observations that include 19 clinical and demographic characteristics of patients suffering from asthma. Data was processed through cleaning and selecting particular features. Then, a 80:20 train and test splitting technique was implemented. The model effectiveness was measured with such parameters as accuracy, precision, recall, F1-score, AUC, sensitivity and specificity. As a result of experiments, KNN classifier provided the highest prediction accuracy and F1-score – 75% and 0.50 correspondingly, which proves its high ability to predict positive cases of asthma. The comparative analysis demonstrates the capabilities and drawbacks associated with each model and shows how essential it is to use suitable machine learning models in order to predict asthma risk. The suggested approach offers a cost-effective and effective way of making decisions that could be useful for healthcare practitioners.