Classification of Mild Cognitive Impairment in Senior Citizens from Blood Samples using Machine Learning

Dementia rates are rising globally, and Malaysia is no different. However, current diagnostic methods for cognitive impairment face challenges regarding cost, accessibility, and precision. In response, this work proposes a new method for the reliable and timely detection of mild cognitive impairment (MCI) in the senior population that uses blood-based biomarkers and machine learning models. Through systematic procedures encompassing feature selection using the filter-based method and various Machine Learning classifiers, findings reveal that 26 selected features significantly contribute to MCI classification, with Logistic Regression performing the best at 64.84% accuracy and an AUC-ROC score of 67.62%. While LR emerged as the top-performing model, it is noteworthy that the attained results fell short of the desired threshold of 70% or beyond. However, despite this shortfall, the outcomes remain promising and encouraging, demonstrating the potential of utilizing blood-based features for MCI diagnosis. Notwithstanding the inherent complexity in using blood samples, characterized by subtle differences in measurements between individuals with normal cognition and those with MCI, this innovative approach could revolutionize early diagnosis and intervention strategies for MCI, thereby improving the well-being and quality of life for affected individuals.