Geriatric depression symptoms coexisting with cognitive decline: A comparison of classification methodologies

•Depressive signs in concurrent cognitive decline are detected through mining EEG resting-state activity.•Random Forest, Random Tree, MLP Network and Support Vector Machines (SVM) are employed for data classification.•Random Forest demonstrated the highest accuracy.•Synchronization features significantly contributed to the decision tree formation.

Geriatric depression is a pathological process that causes a great number of symptoms resulting in limited mental and physical functionality. The computation of oscillatory and synchronization patterns in certain brain areas may facilitate the early and robust identification of depressive symptoms. In this study electroencephalographic (EEG) data were recorded from 34 participants suffering from both cognitive impairment and geriatric depression (mean age 69.81) and 32 control subjects (mean age 70.33). Both groups were matched according to their cognitive status. The study aims at evaluating neurophysiological features of elderly participants suffering from depression and neurodegeneration. The current work focuses on the identification of depression symptoms that coexist with cognitive decline, the correlation of the examined neurophysiological features with geriatric depression combined with cognitive impairment and the investigation of the role of data mining techniques in the analysis of EEG data. The EEG features were estimated through synchronization analysis (Orthogonal Discrete Wavelet Transform). Depressive patterns were detected through data mining techniques. Random Forest, Random Tree, Multilayer Perceptron (MPL Network) and Support Vector Machines (SVM) were employed for data classification. The efficiency of the classifiers varied from 92.42 to 95.45%. Random Forest demonstrated the highest accuracy (95.5%). Both synchronization and oscillatory features contributed to the decision trees’ formation, with the former prevailing. Moreover, synchronization features significantly contributed to the decision trees’ formation. In line with previous neuroscientific findings, synchronization among right frontal–midline anteriofrontal regions showed great correlation with depressive symptoms. Evaluation of the classifiers indicated the Random Forest as being the most robust algorithm. Synchronization of certain brain regions is more indicative of identifying depression symptoms than oscillatory since synchronization features contributed the most to the formation of the classification trees.