Open Access

ARTICLE

Stress Detection of IT and Hospital Workers Using Novel ResTFTNet and Federated Learning Models

Pikkili Gopala Krishna^1,*, Jalari Somasekar²

1 Department of Computer Science and Engineering, Vemana Institute of Technology, Koramangala, Bangalore, 560034, India
2 Department of Computer Science and Engineering, Jain (Deemed-to-be University), Bangalore, 560069, India

* Corresponding Author: Pikkili Gopala Krishna. Email:

Intelligent Automation & Soft Computing 2025, 40, 235-259. https://doi.org/10.32604/iasc.2025.063657

Received 20 January 2025; Accepted 06 March 2025; Issue published 28 April 2025

Abstract

Stress is mental tension caused by difficult situations, often experienced by hospital workers and IT professionals who work long hours. It is essential to detect the stress in shift workers to improve their health. However, existing models measure stress with physiological signals such as PPG, EDA, and blink data, which could not identify the stress level accurately. Additionally, the works face challenges with limited data, inefficient spatial relationships, security issues with health data, and long-range temporal dependencies. In this paper, we have developed a federated learning-based stress detection system for IT and hospital workers, integrating physiological and behavioral indicators for accurate stress detection. Furthermore, the study introduces a hybrid deep learning classifier called ResTFTNet to capture spatial features and complex temporal relationships to detect stress effectively. The proposed work involves two local models and a global model, to develop a federated learning framework to enhance stress detection. The datasets are pre-processed using the bandpass filter noise removal technique and normalization. The Recursive Feature Elimination feature selection method improves the model performance. FL aggregates these models using FedAvg to ensure privacy by keeping data localized. After evaluating ResTFTNet with existing models, including Convolution Neural Network, Long-Short-Term-Memory, and Support Vector Machine, the proposed model shows exceptional performance with an accuracy of 99.3%. This work provides an accurate and privacy-preserving method for detecting stress in hospital and IT staff.

---

Keywords

---