Background: Major Depressive Disorder (MDD) involves complex disturbances in hedonic processing, which contribute to anhedonia—a core symptom of the disorder. Although anhedonia is well recognized, the relative contributions of distinct hedonic components remain poorly understood. Advances in machine learning (ML) provide powerful tools to model high-dimensional data and may clarify the most critical components for differentiating MDD from healthy individuals.
Methods: Sixty-six MDD patients and 249 healthy controls completed the Monetary Incentive Delay task. A voting classifier was trained on the entire dataset to develop an integrated MDD classification model. To evaluate component-level importance, the classifier was applied to feature subsets reflecting different combinations of hedonic components. Individual-level importance was further examined using importance coefficients, stability selection, and statistical tests.
Results: The integrated model demonstrated strong diagnostic performance, achieving an area under the curve (AUC) of 0.806, sensitivity of 78.5%, and specificity of 82.6%. Across all optimal models that achieved the highest AUC (0.806), sensitivity (81.1%), or specificity (90.1%), anticipatory pleasure consistently emerged as a key predictive feature component. Meanwhile, five of the six top-ranked features were also derived from anticipatory pleasure.
Conclusions: These findings underscore the importance of integrating multi-dimensional hedonic processing to classify MDD from healthy individuals and emphasize anticipatory pleasure as a core deficit within hedonic processing of MDD.
Supplementary Information: The online version contains supplementary material available at 10.1186/s12888-026-08013-0.
Keywords: Anticipatory pleasure; Hedonic processing; Machine learning; Major depressive disorder; Monetary incentive delay task; Voting classifier.