Table 1.
Summaries of the methods and outcomes of previous literature reviews
| Method/classifier | Dataset | Performance metric | References |
|---|---|---|---|
| ADHD detection | |||
| Support vector Machine | 50 individuals of 10–12 aged school children in Chile with their oculometry data (such as, gaze position, blink frequency and pupil size) from 22 custom engineered features | 75% accuracy, 77% sensitivity and 74% specificity | [10] |
| Linear discriminant analysis | 5726 college students | 93.1% accuracy | [11] |
| Support Vector Machine | 150 individuals from the national institutes in Boston, USA | 96.5% accuracy | [12] |
| Support Vector Machine | 36 ADHD and 35 normal cases of a private dataset | 74.65% accuracy, 75% sensitivity and 74.29% specificity | [13] |
| Single-channel deep neural network | Multiscale brain connectome data and personal characteristic data of 973 participants from Neuro Bureau ADHD-200 dataset | 82% AUC | [14] |
| Support Vector Machine with Recursive Feature Elimination | 159 structural MRI images of children from Neuro Bureau ADHD-200 dataset | 60.78% accuracy | [15] |
| Extreme learning machine | 776 individuals from 7 to 14 years with their structural MRI data from Neuro Bureau ADHD-200 dataset | 76.19% accuracy | [16] |
| Support Vector Machine | 973 participants including ADHD patients and healthy controls from Neuro Bureau ADHD-200 dataset | 75% accuracy | [17] |
| Boruta based feature selection and support vector machine for identification | Comorbidity-free ADHD individuals with covariable-matched healthy children aged 9–10 chosen from the Adolescent Brain and Cognitive Development study | 64.3% accuracy and 69.8% AUC | [18] |
| Support vector machine | fMRI scan and phenotypic data (age, gender, handedness, IQ, and site of scanning) of 668 participants from the ADHD-200 dataset | 76% accuracy on 2 class and 68.6% on 3 class accuracy | [19] |
| Novel Knowledge Distillation-Based Feature Selection based on neural network | 776 training and 197 testing cases from the ADHD-200 dataset | Average 75% accuracy on different sites | [20] |
| Linear support vector machine | ADHD-200 data set rom Kennedy Krieger Institute (KKI), NeuroImage (NI), New York University Medical Center (NYU) and Peking University (Peking) | 56–81% accuracy on different dataset | [21] |
| Support vector machine-recursive feature elimination | 83 boys aged 7–15 with ADHD, 86 boys with ASD, 125 boys with typical development | 79.3% accuracy | [22] |
| OCD detection | |||
| Logistic regression in neural network | 47 children and adolescents with OCD and 17 children and adolescents without a psychiatric diagnosis, aged 8–17 years | 89% accuracy, 78% sensitivity, 86% specificity, and an 89% AUC score | [23] |
| Gradient enhanced decision trees | 215 observations and 227 features | AUROC score of 78.2%, sensitivity of 73.42%, and specificity of 71.45% | [24] |
| Random Forest | 533 participants of a specialised outpatient clinic | 65% accuracy | [25] |
| Random Forest | 61 adolescents with 46 demographic and clinical baseline variables by Child and Adolescent Psychiatry Research Centre in Stockholm, Sweden | 83% accuracy | [26] |
| Support Vector Machine | 68 drug naïve OCD patients of Chinese Han nationality | 72% accuracy | [27] |
| Ensemble method | 330 Iranian patients with 36 features | 86% accuracy | [28] |
| Support Vector Machine | 296 individuals using 24 baseline variables | 75.4% accuracy | [29] |
| Support Vector Machine | 54 drug naïve Chinese OCD patients with 6 motion parameters | Accuracy 95.37%, sensitivity 96.30%, and specificity 94.44%, | [30] |
| SAD detection | |||
| Alternating decision trees | Children aged 2–5 who visited Duke University Paediatric Primary Care Clinics | 96% accuracy | [31] |