A Prognostic Model to Improve Asthma Prediction Outcomes Using Machine Learning

• Published in: The open bioinformatics journal Vol. 17; no. 1
• Main Authors: M R, Pooja, Ravi, Vinayakumar, Lokesh, Gururaj Harinahalli, Al Mazroa, Alanoud, Ravi, Pradeep
• Format: Journal Article
• Language: English
• Published: 28.06.2024
• ISSN: 1875-0362; 1875-0362
• DOI: 10.2174/0118750362306414240624113350

Purpose The utility of predictive models for the prognosis of asthma disease that rely on clinical history and findings has been on the constant rise owing to the attempts to achieve better disease outcomes through improved clinical processes. With the prognostic model, the primary focus is on the search for a combination of features that are as robust as possible in predicting the disease outcome. Clinical decisions concerning obstructive lung diseases such as Chronic obstructive Pulmonary Disease (COPD) have a high chance of leading to results that can be misinterpreted with wrong inferences drawn that may have long-term implications, including the targeted therapy that can be mistakenly beset. Hence, we suggest data-centric approaches that harness learning techniques to facilitate the disease prediction process and augment the inferences through clinical findings. Methods A dataset containing information on both symptomatic representations and medical history in the form of categorical data along with lung function parameters, which were estimated using a spirometer (with the data basically being quantitative (numerical) in nature) was used. The Naïve Bayes classifier performed comparatively well with the optimized feature set. The adoption of One-Class Support Vector Machines (OCSVM) as an alternative method to sampling data has resulted in the selection of an ideal representation of the data rather than the regular sampling approach that is used for undersampling. Results The model was able to predict the disease outcome with a precision of 86.1% and recall of 84.7%, accounting for an F1 measure of 84.5%.The Area under Curve(AUC) and Classification Accuracy (CA) were evaluated to be 92.2% and 84.7% respectively. Conclusion Incorporating domain knowledge into the prediction models involves identifying clinical features that are most relevant to the process of disease classification using prior knowledge about the disease and its contributing factors, which can significantly enhance the productivity of the models. Feature engineering is centric on the use of domain knowledge within clinical prediction models and commonly results in an optimized feature set. It is evident from the experimental results that using a combination of medical history data and significant clinical findings result in a better prognostic model