Abstract

Mild Cognitive Impairment (MCI) is a prevalent condition whose early detection is critical to preventing progression toward major forms of dementia. This study proposes a supervised learning framework for identifying individuals with suspected MCI, using a dataset of 4,000 simulated cognitive, temporal, and behavioral records. Four classification algorithms—Logistic Regression, Random Forest, SVM, and XGBoost—were trained and evaluated, prioritizing recall to reduce false negatives in clinical settings. Results indicate that ensemble models (XGBoost and Random Forest) achieved recall scores of 75 % and F1-scores above 70 %, outperforming linear methods in precision and class balance. Individuals classified as MCI will be divided into two groups: one undergoing an immersive virtual reality (VR) intervention, and the other receiving conventional treatment. The evolution of both groups will be assessed through pre- and post-intervention statistical analysis; however, this second phase will be addressed in future work. The methodology presented here lays the groundwork for intelligent diagnostic tools that integrate artificial intelligence with immersive environments for future clinical applications. This paper presents the machine learning component of a larger VR-based cognitive assessment system, focusing on the classification models that will drive future adaptive interventions.