Clinical outcome prediction in pediatric respiratory infections using hybrid feature selection and a genetic
Sarlinraj Madhalaimuthu1, Sujatha Radhakrishnan2
1School of Computer Science Engineering and Information Systems, Vellore Institute of Technology, Vellore, Tamilnadu, 632014, India.
Insights
This study introduces a machine learning approach for classifying respiratory datasets, achieving 95.02% accuracy. The method enhances early detection and prediction of clinical progression for respiratory illnesses.
Area of Science:
- Medical Informatics
- Computational Biology
- Data Science
Background:
- Respiratory ailments pose significant health risks, leading to severe complications and mortality if untreated.
- Effective management requires early detection, prevention, and advanced treatment options.
- Machine learning offers potential for improved analysis of complex respiratory datasets.
Purpose of the Study:
- To develop and validate a comprehensive machine learning (ML) method for enhanced investigation and classification of respiratory datasets.
- To improve the prediction accuracy of clinical progression in respiratory diseases.
- To create an interpretable and accurate ML model for healthcare analytics.
Main Methods:
- Data preprocessing and augmentation, including Synthetic Minority Oversampling Technique (SMOTE) for data imbalance.
- Feature selection strategies to identify relevant characteristics and reduce dimensionality.
- Ensemble learning combining Genetic Algorithms, Random Forest, XGBoost, and Gradient Boosting for classification.
Main Results:
- The proposed ML model achieved an overall accuracy of 95.02% in predicting clinical progression.
- The method demonstrated high predictive accuracy while maintaining model interpretability.
- Optimal models including Random Forest, XGBoost, and Gradient Boosting were identified and combined.
Conclusions:
- The developed machine learning framework provides a robust system for classifying respiratory datasets and predicting clinical progression.
- This approach can significantly aid in healthcare analytics for respiratory disease management.
- The high accuracy and interpretability of the model support its application in clinical settings.
Abstract:
Respiratory ailments constitute various pathological conditions affecting the respiratory system, including the airways, pulmonary tissues, and associated structures. When these conditions are left untreated or inadequately managed, they can result in long-term complications, diminished life quality, and higher death rates. To alleviate the strain of respiratory illnesses and promote a more robust population, it is crucial to focus on raising public awareness, facilitating early detection, implementing preventive strategies like immunization, and furthering medical advancements in treatment options. The study presents a comprehensive Machine Learning (ML) method to improve the investigation and classification of respiratory datasets. The technique applies data preprocessing, augmentation, feature selection, genetic algorithms, and ensemble learning techniques on a "Respiratory dataset" and achieves high predicted accuracy while maintaining interpretability. The Synthetic Minority Oversampling Technique (SMOTE) is used to address data imbalance and ensure proper representation of minority class samples. The feature selection module uses various strategies to find relevant characteristics and reduce dimensionality. Machine learning algorithms that are apt for the dataset are employed for predicting the target variable; their performance is measured and analyzed thoroughly. By using Genetic algorithms, Random Forest, XGBoost, and Gradient Boosting are selected as optimal models. The ensemble learning framework combines the 3 optimal models and creates a strong classification system to predict "target variable : Clinical Progression" output. The performance measures of the proposed model achieved an overall accuracy of 95.02% when compared with the existing works and can be applied in healthcare analytics.
