چكيده به لاتين
In the past decade, the use of artificial intelligence and machine learning in research related to multiple sclerosis (MS) has led to significant advancements. These technologies have enhanced researchers' ability to analyze complex data, enabling early diagnosis and improving treatment processes. This study aims to examine the application of fuzzy expert systems and medical data mining in predicting specific diseases such as MS. Fuzzy expert systems, due to their capability to handle uncertain data and process complex information, can contribute to higher accuracy in disease diagnosis and prediction. Additionally, ensemble-based data mining techniques help identify hidden patterns in patient data, thereby improving prediction accuracy.
This research utilizes data from patients with clinically isolated syndrome (CIS) to predict the likelihood of its progression to definitive multiple sclerosis. The dataset includes demographic features, clinical symptoms, and MRI results. For disease prediction, a hybrid approach was employed, combining a rule-based fuzzy expert system with machine learning algorithms, including XGBoost, AdaBoost, and Random Forest. The fuzzy expert system was designed based on a set of medical rules, taking inputs such as age, oligoclonal bands, visual evoked potential, and MRI findings to calculate the probability of disease progression. The machine learning models were trained using parameter optimization techniques and compared against each other.
The evaluation results of the fuzzy expert system indicated that, in most cases, the model's predictions aligned with actual outcomes; however, there remains a need for accuracy improvement in some instances. Furthermore, the results demonstrated that XGBoost outperformed the other two algorithms across all evaluation metrics, including accuracy, precision, recall, and F-score. The accuracy of XGBoost was 0.8000, surpassing AdaBoost (0.7818) and Random Forest. Additionally, XGBoost exhibited superior precision (0.8000), recall (0.7984), and F-score (0.7989) compared to the other methods.
