Developing A Self-Learning Expert System For Respiratory Diseases Diagnosis And Treatment

Abstract

Globally, there is a significant health burden associated with respiratory diseases for instance, Chronic obstructive pulmonary disease (COPD) affects over 200 million individuals, asthma impacts 235 million people, and tuberculosis (TB) afflicts 8.7 million annually. Detecting these diseases at an early stage is crucial for effective treatment. However, the availability of experts, especially in developing nations, is often limited. Therefore, the main objective of this research is to develop a self-learning expert system for respiratory diseases diagnosis and treatment. The research utilized a design science research approach to gain insights into the challenges within the field and create a model as a solution. In this study, a primary focus was on employing classification techniques within data mining tasks to extract representative cases from the collected data. To determine the optimal model and select the most effective data mining classification algorithm, three experiments were carried out using J48, PART and JRIP classification algorithms. The data mining algorithm achieved accuracy of 94.5%, 92.5% and 90% for PART, J48 and JRIP respectively. This system aims to identify and offer insights into the diagnosis and treatment of common respiratory diseases. In order to develop the model the researcher collected both implicit and explicit knowledge from various sources, including domain experts, medical literature, research papers, clinical guidelines and patient records. Then organized the acquired knowledge, selected the most important attributes and discarded unnecessary ones by analysing the patient records. The relevant information was transferred to an Excel sheet and saved as a CSV file. The researcher utilized the WEKA 3.9 data mining software to pre process and classify the data using algorithms such as J48, PART, and JRIP. The performance of these algorithms was evaluated using a 10-fold cross-validation method and an 80/20% data split. Finally, the PART rules were represented in a structured format suitable for computer-based systems using rule-based representation techniques and integrated with a node to run the main interface. The prototype is evaluated using system testing and user acceptance testing. System testing performed in terms of recall, precision and F-measure registered 90.9%, 90.9% and 90.5% respectively. User acceptance testing also performed by involving domain experts and an average of 84% acceptance was achieved. To enhance the widespread adoption of knowledge-based systems by the general public it is recommended to add more local languages to avoid language barriers.