Intelligent Web Application based on RF and CNN for the Multiclass Classification of Cognitive Impairment in Older Adults

Cognitive impairment remains a growing public health challenge in aging populations, particularly in regions with limited access to specialized diagnosis tools. This study proposed an intelligent web-based system for the multiclass classification of cognitive impairment in older adults, based on structured clinical data and magnetic resonance imaging (MRI). The system was designed to offer a practical, low-cost alternative to traditional methods, enabling healthcare professionals to make accurate early diagnoses using accessible data. The platform includes two predictive modules: one based on Random Forest using clinical and functional data, and another based on Convolutional Neural Networks (CNN) for MRI classification. To assess performance, we trained and tested models on validated datasets, optimizing hyperparameters with Optuna. The Random Forest model achieved an accuracy of 81.8% and an F1 score of 0.868 for classifying patients into Control or MCI+Dementia, while the CNN model reached 99% across all metrics using only MRI scans. The system also includes a visualization dashboard to facilitate diagnostic decisions. These results suggest that machine learning models can complement clinical judgment and support timely diagnosis in resource-limited settings. The proposed tool is promising for scalable deployment and can be extended with multimodal data in future research.