Technostress prediction among students in Finland using machine learning

Abstract

Technostress is a condition that is becoming more widely acknowledged as a hindrance to both student mental health and successful learning as digital technology permeates education. Given the increasing integration of digital tools and platforms into educational settings, it is critical to comprehend the elements that lead to technostress. We explored the factors and prediction models of technostress with 202 participants (including 38 pilot research participants from several Finnish institutions). We used a combined strategy that included machine learning (ML) techniques to predict and categorize technostress levels, statistical analysis to find parameters related to technostress, and the Shapiro-Wilk test to optimize questionnaire components. Our research showed that whereas coping flexibility is adversely connected with technostress, techno-overload and techno-invasion are strongly and positively correlated. Testing the hypothesis revealed that every factor—apart from privacy—significantly influences the measurement of technostress. In our comparative study, we found that our Neural Network-based multi-layer perceptron (MLP) classifier with principal component analysis (PCA) outperformed conventional ML algorithms, achieving a 71% classification accuracy as well as over 70% precision, recall, and F1-score. These findings imply that our model successfully detects students who are at risk of technological stress, offering educational institutions a useful tool for creating support and intervention plans. The results of our study underscore the vital necessity of focused interventions aimed at reducing students' technological stress. This can help to shape educational policies and mental health approaches that improve students' overall health and academic achievement. Furthermore, machine learning model developers can learn a lot from the proven performance of the MLP classifier with PCA in managing small datasets. This includes being encouraged to use advanced neural network techniques and dimensionality reduction for robust performance in similar constrained data scenarios.