Improving patient’s medical history classification using a feature construction approach based on situation awareness and granular computing

Abstract

Healthcare decision support systems aid physicians in disease classification by analyzing patients’ medical histories to suggest preliminary diagnoses. As physicians largely base their analysis on anamnesis, integrating this process into an automated recommendation system can expedite decision-making and transition to relevant clinical investigations, thus enhancing efficiency in diagnosing potential pathologies. In this research, an innovative method for feature construction is introduced, drawing on the concepts of Situation Awareness and Granular Computing. The aim of this method is to enhance the performance of out-of-the-box classification algorithms used in machine learning. The approach is specifically tailored to mimic physicians’ cognitive processes when analyzing a patient’s medical history, resulting in the generation of new, information-dense features that can be used for classification tasks. By employing this strategy, a deeper comprehension of the data can be achieved, as well as a more precise categorization of anamneses in relation to possible medical conditions. To authenticate the efficacy of the proposed technique, three major disease categories, namely cardiac, gastrointestinal, and thyroid, were considered. The dataset comprised 1213 medical histories. The experimental results indicate that the study’s six classifiers attained a balanced accuracy exceeding 90%. Among these, the SVM classifier demonstrated the highest balanced accuracy at 93%. Overall, the proposed approach resulted in an average increase of 16 percentage points in balanced accuracy, representing an improvement over the traditional methods commonly employed in machine learning. This approach could be integrated into a clinical decision support system, aiding physicians in accurately identifying necessary investigations and expediting diagnosis.