A Machine Learning Approach for Predicting Clinical Outcomes in Patients with Autoimmune Encephalitis

Autoimmune encephalitis (AE) is an uncommon but severely debilitating neuroinflammatory disorder that can result in significant neuropsychiatric issues and prolonged disability. The ability to predict patient outcomes accurately and promptly poses a considerable clinical challenge, largely due to the disease's variability and the intricate interactions among clinical, laboratory, and imaging factors. This research aimed to create a comprehensive machine learning-based predictive model designed to help clinicians forecast AE outcomes with a high degree of precision. A dataset obtained from Kaggle, which is publicly accessible, was employed, featuring diverse patient information such as demographic data, diagnostic indicators, and characteristics derived from neuroimaging. The preprocessing of data encompassed meticulous management of missing values, normalization processes, and the alleviation of class imbalance through the application of the Synthetic Minority Oversampling Technique (SMOTE). The predictive model was developed using Python and the scikit-learn library, evaluating various algorithms including Random Forest, Gradient Boosting, and Support Vector Machines. To enhance the model's generalizability, hyperparameter tuning was conducted via grid search in conjunction with stratified cross-validation. The assessment of performance was based on metrics such as accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC-ROC). Among the evaluated models, the Random Forest classifier demonstrated superior performance, attaining an accuracy of 91.4%, an F1-score of 0.89, and an AUC-ROC of 0.94, which emphasizes its robust discriminative ability. These results point to the promise of machine learning techniques, when trained on actual datasets, to improve decision-making in adverse event management, thereby facilitating more tailored and evidence-based clinical strategies.