Citation
Haque, Radiah (2024) Machine learning techniques for asthma weather-based healthcare. PhD thesis, Multimedia University. Full text not available from this repository.Abstract
Machine Learning (ML) is an innovative scientific approach that uses various techniques (e.g., regression, classification, and optimisation) for prediction and pattern recognition. Recently, ML techniques have been successfully used for healthcare applications, especially for detecting diseases and predicting chronic disease exacerbation. Nevertheless, it can be observed that there is still a challenge in using ML techniques to identify the relationships between weather and diseases to assist Weather-Based Healthcare (WBH) and offer accurate predictions of disease exacerbation based on personalised weather triggers. WBH refers to self-management of chronic diseases that are affected by weather, such as asthma. Studies show that weather triggers (e.g., temperature and humidity) often lead to asthma exacerbation and acute asthma attacks. Nonetheless, weather impact is specific to individual asthmatic patients due to their lung performance, which varies among patients based on their demographic characteristics (e.g., age and gender). As such, personalisation is significant to recognise the potential weather attributes that can trigger individual patients’ asthma. In this case, ML techniques can be used to develop an intelligent system for effective asthma self-management. In recent years, there have been attempts to develop Mobile Health (mHealth) applications for asthma self-management. However, until now, no solution for effective asthma self-management exists that has been widely adopted by users or integrated into primary asthma care records. This is because there is a lack of ML models that can offer accurate predictions of asthma exacerbation based on demography and weather and provide tailored feedback to asthmatic patients. This study aims to apply ML techniques for predicting asthma exacerbation based on personalised weather triggers to enhance asthma self- management and assist WBH. With the aim of integrating weather, demography, and asthma tracking, an mHealth application, namely Weather Asthma (WEA), was developed where users conduct the Asthma Control Test (ACT) to identify the severity and chances of asthma exacerbation. The asthma dataset consists of panel data from asthmatic patients who suffer from mild (e.g., cough-variant asthma) to extreme conditions (i.e., acute asthma), and includes around 3000 ACT scores as the target output. Moreover, the dataset contains ten input features with five weather variables (temperature, humidity, air pressure, UV index, and wind speed) and five demographic variables (age, gender, outdoor job, outdoor activities, and location). Subsequently, two predictive models were developed and applied on the asthma dataset: Feedforward Deep Neural Network (FDNN) model for conducting regression analysis and Multiclass Classification Neural Network (MCNN) model. Using the FDNN model, an accuracy of 88% was achieved with Mean Absolute Error (MAE) = 1.23, Mean Squared Error (MSE) = 2.96, Root Mean Squared Error (MSE) = 1.72, and Explained Variance Score (EVS) = 0.87. It was recognised that, for effective asthma self-management, the prediction error must be in the acceptable loss range (error < 0.5). Hence, an optimisation process is proposed to reduce the error rates and increase the prediction accuracy by utilising data scaling and hyperparameter tuning techniques. In effect, normalisation using Z-score and Min-Max techniques is applied for scaling both input and output variables. The experimental results reveal that the normalisation technique improves the stability of the FDNN model, which also enhances the performance of the Adam optimiser. Furthermore, in order to tune the hyperparameters and identify the optimal structure of FDNN, the Fragmented Grid Search (FGS) optimisation algorithm is proposed. The results show that the FGS method is able to tune several hyperparameters of the FDNN model with much less computing time (≈ 20 minutes) than the standard grid search method. Consequently, the optimised FDNN model achieved an accuracy of 96%, with MAE = 0. 0409, MSE = 0.0038, RMSE = 0.0618, and EVS = 0.95. Subsequently, the FGS optimisation algorithm helped to build the MCNN model which provides predictions with a high accuracy rate 96%, precision = 0.9642, recall = 0.9604, F1 score = 0.9649, and ROC AUC is 99%. The empirical investigation in this study recognises the potentials of ML techniques to identify the correlation patterns among asthma, weather, and demographic variables, which interplay when predicting asthma exacerbation based on personalised weather triggers. The findings will help to integrate the optimised predictive model into mHealth solutions to build intelligent systems for effective asthma self-management and WBH.
Item Type: | Thesis (PhD) |
---|---|
Additional Information: | Call No.: Q325.5 .R33 2024 |
Uncontrolled Keywords: | Machine learning |
Subjects: | Q Science > Q Science (General) > Q300-390 Cybernetics |
Divisions: | Faculty of Computing and Informatics (FCI) |
Depositing User: | Ms Nurul Iqtiani Ahmad |
Date Deposited: | 04 Jul 2025 02:02 |
Last Modified: | 04 Jul 2025 02:14 |
URII: | http://shdl.mmu.edu.my/id/eprint/14215 |
Downloads
Downloads per month over past year
![]() |