Beyond classification metrics: a psychometric-aware benchmark for data augmentation in imbalanced student mental health surveys - Takeaways - MDSpire

Beyond classification metrics: a psychometric-aware benchmark for data augmentation in imbalanced student mental health surveys

  • By

  • Chen Shao

  • Shengnan Qiao

  • Ang Li

  • Shipeng Liu

  • Yanglong Chen

  • Yuzhe Tan

  • Zhibo Liang

  • Xuanming Si

  • June 15, 2026

  • 0 min

Share

  • 1

    Machine learning can enhance depression screening in students but faces challenges of class imbalance and psychometric validity in synthetic data.

  • 2

    The study benchmarks eight data-augmentation strategies for their classification utility and construct validity on mental health datasets.

  • 3

    PCT-GAN outperformed other methods in maintaining psychometric validity while improving classification performance on imbalanced datasets.

  • 4

    SMOTE variants excelled in large datasets, while PCT-GAN showed significant gains in smaller, imbalanced datasets like PHQ-9.

  • 5

    The findings suggest using SMOTE for transient training and PCT-GAN for sharing synthetic data in psychometric research.

Original Source(s)

Beyond classification metrics: a psychometric-aware benchmark for data augmentation in imbalanced student mental health surveys

  • by Chen Shao, Shengnan Qiao, Ang Li, Shipeng Liu, Yanglong Chen, Yuzhe Tan, Zhibo Liang, Xuanming Si

  • June 15, 2026

Related Content

Impact of rehabilitation trajectory on affective and cognitive impairment after intracerebral hemorrhage: a cohort study

Comparative effectiveness of non-pharmacological interventions on depression and anxiety in aging populations: a systematic review and network meta-analysis of randomized controlled trials

Efficacy and safety of sini powder combined with Xiaoxianxiong decoction for adolescent depression with liver qi stagnation and phlegm-heat obstruction syndrome: a randomized controlled trial