An App for Detecting Bullying of Nurses Using Convolutional Neural Networks and Web-Based Computerized Adaptive...

Area of Science:

• Nursing Research
• Artificial Intelligence in Healthcare
• Occupational Health Psychology

Background:

• Workplace bullying significantly impacts mental health, necessitating accurate measurement tools.
• Existing methods for assessing workplace bullying lack advanced classification and adaptive testing capabilities.
• No prior studies have integrated Convolutional Neural Networks (CNN) with Computerized Adaptive Testing (CAT) for workplace bullying assessment.

Purpose of the Study:

• To develop a novel mobile application for the automatic detection and classification of bullying levels among nurses.
• To integrate CNN with online Rasch Computerized Adaptive Testing (CAT) for early-stage assessment of nurse bullying.
• To enhance the accuracy and efficiency of bullying assessment in the nursing profession.

Main Methods:

• Recruited 960 nurses and administered the 22-item Negative Acts Questionnaire-Revised (NAQ-R).
• Employed k-means clustering for unsupervised classification and CNN for supervised learning to categorize bullying severity.
• Developed a predictive model using 70:30 training-testing data split, estimating 69 parameters and incorporating Rasch CAT.

Main Results:

• The 22-item CNN model achieved high accuracy (94% overall) in classifying bullying levels.
• Specific accuracies reached 99% for lower and 83% for upper bullying severity groups (AUCs 0.99 and 0.94, respectively).
• The predictive model demonstrated strong performance, with 95% accuracy in the training set and 97% in the testing set.

Conclusions:

• The 22-item CNN model combined with Rasch online CAT significantly improves NAQ-R assessment accuracy for nurse bullying.
• A functional app was successfully developed for nurses to self-assess workplace bullying.
• The study recommends the app for future application in early-stage workplace bullying detection and intervention.