Skip to main content
Fig. 1 | BMC Psychiatry

Fig. 1

From: Transdiagnostic symptom subtypes across autism spectrum disorders and attention deficit hyperactivity disorder: validated by measures of neurocognition and structural connectivity

Fig. 1

The characters of the unsupervised classification analysis. a The dendrogram for hierarchical clustering. The y-axis represents the distance between clusters. Colors represent the three-cluster solution chosen. b The gap statistic, calculated for cluster counts from 1 to 15. The line charts show the optimal cluster number based on the gap statistic values and the red circles indicate the optimal cluster number. In this context, the most optimal cluster number was 3. c The distribution of adjusted Rand scores visualized in a histogram. To evaluate the stability of the clustering solution, we repeated the clustering analysis in randomly selected subsamples (each containing 80% of the subjects) for 1000 times. In each of the 1000 subsamples, the remaining 20% subjects left out were assigned to clusters using linear discriminant analysis classifiers. These two samples combined to form a complete cluster solution. We then tested the stability of clustering over the 1000 subsamples by calculating an adjusted Rand score, which represent the similarity between each clustering solution compared to the original clustering solution. The average adjusted Rand score was 0.66 (min 0.21, max 0.79). d. Linear discriminant analysis. Linear discriminant analysis is a supervised classification method that constructs a predictive model to evaluate the group membership, based on the theory of Bayes formula. Thus, these new variables are called discriminants functions (DF’s) that provide the best discrimination between the groups

Back to article page