DESIGN AND IMPLEMENTATION OF MACHINE LEARNING TECHNIQUES FOR MALARIA INCIDENCE AND TUBERCULOSIS PREDICTION

DESIGN AND IMPLEMENTATION OF MACHINE LEARNING TECHNIQUES FOR MALARIA INCIDENCE AND TUBERCULOSIS PREDICTION

Abstract

The increasing burden of infectious diseases in developing countries has highlighted the need for intelligent healthcare solutions capable of supporting timely diagnosis, disease surveillance, and evidence-based decision-making. Recent advances in artificial intelligence, particularly machine learning, have created new opportunities for improving disease prediction and public health management by analyzing large volumes of health-related data. This study presents the design and implementation of machine learning models for predicting malaria incidence and drug-resistant tuberculosis (TB), with the aim of enhancing healthcare decision support systems and improving disease control strategies in Africa.

Malaria remains one of the leading causes of illness and death across sub-Saharan Africa despite significant global intervention efforts. The region continues to account for the overwhelming majority of malaria infections and fatalities, with countries such as Nigeria, the Democratic Republic of Congo, Tanzania, Burkina Faso, Mozambique, and Niger experiencing the highest disease burden. The transmission of malaria is strongly influenced by environmental and climatic conditions, including temperature, rainfall, humidity, and seasonal weather variations. However, the impact of these climatic factors differs across geographical regions, making localized prediction models essential for effective disease surveillance and prevention.

To address this challenge, this research develops a machine learning-based malaria prediction model using the Extreme Gradient Boosting (XGBoost) algorithm. XGBoost is selected because of its high predictive accuracy, computational efficiency, scalability, and ability to process large datasets through parallel learning techniques. The proposed model analyzes historical malaria records alongside environmental and climatic variables to classify malaria incidence levels and predict potential outbreaks. The resulting prediction system serves as an early warning mechanism that enables healthcare authorities, governments, and policymakers to implement preventive interventions before outbreaks escalate. By providing location-specific predictions, the model supports strategic planning, efficient allocation of healthcare resources, and improved malaria control programs in endemic regions.

In addition to malaria prediction, this study focuses on improving the diagnosis of drug-resistant tuberculosis (DR-TB), one of the most significant global public health challenges. Tuberculosis continues to affect millions of people annually, while multidrug-resistant strains have made disease management increasingly complex. Conventional diagnostic procedures for drug-resistant tuberculosis often require specialized laboratory facilities, involve invasive testing methods, consume considerable time, and depend heavily on experienced medical professionals. These limitations can delay diagnosis and treatment, thereby increasing disease transmission and reducing patient survival rates.

To overcome these challenges, the research applies a data mining approach using the Frequent Pattern Growth (FP-Growth) algorithm to identify hidden relationships among clinical symptoms, patient characteristics, and diagnostic indicators associated with drug-resistant tuberculosis. The discovered association patterns are subsequently integrated into a logistic regression classification model that predicts whether a patient belongs to a drug-resistant or non-drug-resistant tuberculosis category. This hybrid approach enables the extraction of valuable clinical knowledge from healthcare datasets while improving diagnostic efficiency and supporting faster clinical decision-making.

The developed tuberculosis prediction system functions as an intelligent knowledge-based decision support tool capable of assisting healthcare professionals in identifying high-risk patients at an early stage. By uncovering meaningful associations among symptoms and patient data, the system contributes to improved diagnosis, personalized treatment planning, and enhanced disease management. Furthermore, integrating machine learning into tuberculosis diagnosis reduces dependence on manual interpretation and facilitates more consistent clinical assessments.

To assess the effectiveness of the proposed prediction models, comprehensive performance evaluation was conducted using widely accepted machine learning metrics, including classification accuracy, precision, recall, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). In addition, the Akaike Information Criterion (AIC) was employed to compare competing predictive models and identify the most suitable algorithms for each disease prediction task. Experimental results demonstrated that the proposed XGBoost-based malaria prediction model and the FP-Growth-enhanced Logistic Regression tuberculosis prediction model consistently outperformed several conventional machine learning approaches across the selected evaluation metrics.

The findings of this research demonstrate the effectiveness of machine learning techniques in supporting public health surveillance, clinical diagnosis, and healthcare decision-making. The proposed models provide accurate and reliable predictions that can assist physicians, epidemiologists, healthcare institutions, and government agencies in implementing timely interventions, improving disease monitoring, optimizing resource allocation, and strengthening disease prevention programs. Moreover, the developed framework establishes a scalable foundation for integrating artificial intelligence into modern health informatics systems, particularly in resource-constrained environments where rapid and accurate decision-making is essential.

Overall, this study contributes to the advancement of intelligent healthcare technologies by demonstrating how machine learning can be effectively applied to address two of Africa's most significant infectious diseases. The proposed decision support models have the potential to improve disease surveillance, facilitate early diagnosis, reduce mortality rates, and enhance evidence-based healthcare planning, thereby supporting the achievement of sustainable public health outcomes.