A Validated Technostress Instrument and Hybrid Fuzzy Trapezoidal Neural Network for Digital Behavior-Based Stress Classification
Year: 2026
Type: article
Abstract: Excessive smartphone use among university students is associated with technostress; however, questionnaire-based machine learning studies may report inflated performance because of data leakage and rarely assess whether model differences are statistically meaningful. This study develops a 27-item technostress instrument for smartphone users and evaluates a Hybrid Fuzzy Trapezoidal–Artificial Neural Network (Fuzzy-ANN) for stress classification. The instrument covers digital behavior, technostress stressors based on the Tarafdar framework, and psychological symptoms. The four-item Perceived Stress Scale (PSS-4) was used solely as an external criterion for assigning labels. Content validity was assessed by three experts (CVI = 0.904), while reliability was evaluated using responses from 849 participants, producing Cronbach’s alpha values of 0.73–0.89. Respondents were classified as Not Stressed (PSS-4 ? 5) or Stressed (PSS-4 ? 6), based on published PSS-4 normative data. Questionnaire items were fuzzified using trapezoidal membership functions, and both models employed an identical multilayer perceptron architecture with 512 and 256 units. Stratified five-fold cross-validation was applied, with SMOTE and fuzzification parameters fitted exclusively on each fold’s training data. Fuzzy-ANN achieved 76.56% accuracy and 63.91% balanced accuracy, compared with 75.85% and 63.83%, respectively, for ANN; this difference was not statistically significant (p = 0.778). Both models exhibited generalization gaps exceeding 22 percentage points, indicating overfitting in relation to dataset size and class distribution. A data-leaking evaluation protocol inflated balanced accuracy by 24.43 percentage points. Therefore, the principal methodological contribution is the anti-leakage evaluation protocol, which provides a more reliable basis for questionnaire-based machine learning classification research in similar applied settings
Authors: Munawaroh Munawaroh, Abdul Fadlil, Imam Riadi
Institution: Universitas Ahmad Dahlan
Language: English
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W7225905591
