Disease prognosis using machine learning algorithms based on new clinical dataset

Authors : Melike ÇOLAK, Talya TÜMER SİVRİ, Nergis PERVAN AKMAN, Ali BERKOL, Yahya EKİCİ

Pages : 52-68

Doi:10.33769/aupse.1215962

View : 51 | Download : 74

Publication Date : 2023-06-03

Article Type : Research Paper

Abstract :Today, artificial intelligence-based solutions are produced to facilitate human life in almost every field. The healthcare sector is one of the sectors which took advantage of these solutions. Due to reasons such as the world’s ever-expanding population, ongoing epidemics, and the emergence of new disease types, it is becoming increasingly difficult for a patient to benefit from health services quickly and to make an accurate diagnosis. At this juncture, artificial intelligence reduces the patient density in hospitals, enables patients to access accurate information, and allows medical students to practice by seeing new cases. In this study, a new and reliable dataset was created with disease information obtained from various sources under the supervision of a specialist medical doctor. Then, new patient histories were added to the dataset used in the previous study, the experiments were repeated with the same algorithms, and the accuracy score comparison was presented. The created dataset includes 2006 unique patient histories, 358 symptoms, and 141 diseases and we think it will be a valuable dataset for researchers who make developments using machine learning in the field of healthcare. Various machine learning algorithms have been used in the training process to predict diseases belonging to different branches of medicine, such as diabetes, bronchial asthma, and covid. Besides, Support Vector Machine, Naive Bayes, K-Nearest Neighbors, Multilayer Perceptron, Decision Tree, and Random Forest algorithms, we also studied popular boosting algorithms such as XGBoost and LightGBM. All algorithms were validated with cross-validation and performance comparisons were made with different performance metrics such as accuracy, precision, recall, and f1-score. It is also the first study to achieve an accuracy score of 99.33% with a dataset that involves a greater number of diseases than the datasets used in the studies examined.
Keywords : Healthcare symptom checker, clinical decision support systems, machine learning