A Comparative Analysis of Machine Learning and Deep Learning Models Incorporating Attention Mechanisms for Asthma Prediction

Abstract

Asthma is a respiratory disease that affects millions of people globally. Early detection can facilitate timely intervention to prevent complications and mortality. If asthma is predicted early and accurately, it will help in timely intervention and effective management. But the conventional method of asthma diagnosis often struggles to understand complex relationships between patient symptoms, triggers, and demographic factors, leading to less accurate predictions. To address this gap, this study carried out a comparative analysis of machine learning and deep learning- based asthma prediction models, introducing the attention mechanism to enhance the deep learning model. With the use of the Kaggle dataset, this study trained three machine learning algorithms, which include a decision tree, a random forest, Extreme Gradient Boosting (XGBoost), and a deep learning Bidirectional Long Short-Term Memory (Bi-LSTM) with an attention layer. The method utilized in this study follows a unique pipeline, which includes the data collection, data preprocessing, exploratory data analysis, model development, model training, and model evaluation based on accuracy, precision, recall, and F1-scores. During the data preprocessing, the study observed data imbalance in the target variable, and the Synthetic Minority Oversampling Technique (SMOTE) was applied to balance the dataset. The result from this study experiment shows that the Bi-LSTM + Attention model outperformed all the compared machine learning models across all evaluation metrics used, by achieving the highest accuracy of 96%, while XGBoost, random forest, and decision tree achieved accuracies of 95% 94%, and 93%, respectively. The attention mechanism introduced to the BiLSTM model in this study significantly improved the deep learning model’s ability to focus on the most relevant features, enhancing interpretability and clinical trust. The findings demonstrate that the attention-enhanced Bi-LSTM model can serve as a reliable component of a clinical decision support system, assisting healthcare practitioners in early asthma detection and risk stratification, thereby enabling timely intervention and improved patient management.