Evaluation of Five Classifiers for Depression Episodes Detection

Abstract

Depression is a mental disorder manifested through a set of psychological and physical symptoms, such as the presence of sadness, apathy, hopelessness and irritability, among others. According to the World Health Organization (WHO), depression is affecting more than 300 million people worldwide, presenting a prevalence between 3 and 21%. One of the main problems of this high prevalence is the incorrect classification of patients, since many cases are false positive and false negative diagnoses. In this work it is proposed the study of the behavior of five different classification techniques, random forest (RF), conditional inference trees (cTree), K-nearest neighbor (K-NN), support vector machine (SVM) and Naïve Bayes, to identify depressive states through the motor activity of patients contained in the Depresjon dataset. The activity of this dataset is acquired through the smart watch "Actigraph", based on actigraphy. The evaluation of these classification techniques is finally performed in terms of sensitivity, specificity, the receiver operating characteristic (ROC) curve and area under the curve (AUC), to know their performance to automatically detect depressive patients. The results shown values of sensitivity, specificity and AUC, statistically significant, specially for the RF method, which presents sensitivity = 0.8148, specificity = 0.8158 and AUC = 0.8314. Therefore, it is concluded that these classifiers are able to distinguish patients with depression from controls, based on their motor activity, allowing the development of a non-invasive diagnosis tool to support specialists in the correct diagnosis of depression.

Wearable Artificial Intelligence for Anxiety and Depression: Scoping Review

“…Type of publication, n (%) [19,21,23,[25][26][27][28][29][30]34,[38][39][40][41][42][43][44][45][46][48][49][50][51][52][53][54][56][57][58][59][60][61][64][65][66][69][70][71][73][74][75][77][78][79]81,82,84,86,87] 49 (71) Journal article [20,22,24,…”

“…Country of publication, n (%) [24,25,30,31,37,41,42,50,[54][55][56]59,61,66,74,76,77,80,[83][84][85] 21 (30) United States [34,58,65,[69][70][71]86] 7 (10) Mexico [20,32,35,36,43,47] 6 ( 9) Norway [29,38,48,72,78] 5 (7) United Kingdom [26][27][28]46,60] 5 (7) South Korea [33,63,67,79] 4 (…”

“…AI category [19,20,23,25,26,31,33,34,37,39,40,42,46,[49][50][51][52][53][54][55][56][57][58][59][60][61][63][64][65][66][67][69][70][71]73,[75][76][77][78][79][80][81]83,84,86,87] 46 (67) ML a [24,29,32,44,47,62,82...…”

“…Ground truth assessment e [19,20,32,[34][35][36]42,43,47,48,62,65,[68][69][70][71]86] 17 (25) MADRS f [23,24,27,28,30,38,52,53,59,60,73,77,83 [27,29,33,37,40,41,45,46,[49][50][51]54,57,58,63,64,66,69,75,76,78,…”

“…l The number of studies does not add up, as most studies used >1 performance measure. [22,23,28,37,41,44,45,51,58,[60][61][62]70,73] 168,022.5 (428,843.2; 168-1,570,144) Data set size, mean (SD; range) Data set source, n (%) [19,20,28,31,32,34,36,42,43,47,48,62,65,[68][69][70][71]74,86] 19 (28) Open [21][22][23][24][25][26][27]29,30,33,35,[37][38][39][40][41]…”