We evaluated the performance of automatic machine learning classifiers in distinguishing healthy controls from patients with temporal lobe epilepsy and hippocampal sclerosis (TLE-HS), based exclusively on structural (DTI) and functional (fMRI) connectomes. METHOD: Connectomes were derived from DTI and fMRI of 49 TLE-HS patients (23 right) and 47 healthy controls scanned on the same 3T system. Connectivity matrices (80 cortical and subcortical regions) were generated using the FreeSurfer v6.0 wparc atlas. Structural edges were defined by normalized mean FA values; functional edges by Pearson correlations of BOLD signals (Fisher-transformed). Classification models were trained with SVMc, logistic regression, and linear discriminant analysis (LDA) using the NBS_predict toolbox with 10-fold cross-validation (10 repetitions, p = .01, 5 hyperparameter steps, 500 permutations). RESULTS: Patients’ mean age was 35.2 (±8y) vs. 37.5 (±11) in controls; no gender differences (p > .05). Epilepsy duration (5–50y) and seizure frequency were similar by side. For fMRI connectomes, LDA performed best (accuracy 0.67, AUC 0.679, sensitivity–specificity 0.65–0.70). Logistic regression (0.50) and SVMc (0.47) performed worse. For DTI connectomes, LDA again outperformed (accuracy 0.68, AUC 0.682). CONCLUSIONS: Connectome-based predictive models using LDA consistently detected TLE-HS, suggesting potential diagnostic value, though replication in larger cohorts is required.