Intelligent System for Real Time Type and Severity Classification of Malaria using Ensemble Learning Approach
Abstract
This research aimed to develop a machine learning-based system for classifying the severity of malaria, with a focus on cerebral malaria. The dataset was sourced from various healthcare records and features symptoms such as seizures, altered mental status, headache, vomiting, and focal neurological deficits, along with household burden indicators like financial stress and caregiver burden. The model was trained on a large dataset, which was split into training and testing sets to evaluate the performance of the classifier. The Random Forest classifier built in this research utilized multiple decision trees to classify malaria severity as either Cerebral Malaria or Non-Cerebral Malaria. The training process involved fitting the model to the preprocessed data and optimizing it to accurately predict the severity class based on the given symptoms and factors. Upon model evaluation, the results showed an impressive accuracy score of 0.97, indicating a high level of precision in classifying malaria severity. The classification report revealed macro average scores for precision, recall, and F1 score as 0.94, 0.91, and 0.92 respectively. The analysis revealed that 87.2% of the malaria cases were classified as non-cerebral malaria, with only 12.8% falling under cerebral malaria. This distribution aligned with the high prevalence of non-cerebral malaria in most malaria-endemic regions, supporting the focus of the model on differentiating these two classes. Key features, such as seizures, altered mental status, and focal neurological deficits, emerged as the most influential in determining malaria severity. These findings are critical for understanding the psychological and financial toll malaria places on families and communities. The outcomes of this research demonstrate the potential of machine learning models, particularly Random Forest, in improving the diagnosis and management of malaria.
