Machine Learning Approach Using LLMs for Early Detection of Nomophobia in University Students

Nomophobia, or "No Mobile Phone Phobia," is increasingly prevalent among university students, often manifesting as anxiety, stress, and academic difficulties. With smartphones deeply integrated into daily life, students may experience discomfort or panic when separated from their devices, negatively affecting mental well-being and learning outcomes. Early and accurate detection is therefore crucial; however, traditional survey-based assessments are often time-consuming, subjective, and limited in predictive capability. This research proposes a machine learning framework leveraging Large Language Models (LLMs) for multi-class classification of nomophobia severity in students. A structured survey dataset of 113 responses was collected and preprocessed through rigorous cleaning, anonymization, and preservation of linguistic context. The dataset was stratified into training, validation, and testing subsets to ensure robust evaluation and prevent overfitting. Multiple state-of-the-art LLMs including Qwen 3, Mistral 7B, Gemma 2B, and Phi-3 Mini were fine-tuned using Quantized Low-Rank Adaptation (QLoRA) for optimal classification performance. Experimental results showed Qwen 3 achieved the highest performance with 92.25% accuracy, 92.27% precision, 92.25% recall, and a 92.19% F1-score. Mistral 7B and Gemma 2B also performed strongly, with accuracies of 91.25% and 91.00%, respectively, while the lightweight Phi-3 Mini attained a competitive 90.25% accuracy. These results demonstrate the effectiveness of fine-tuned LLMs in predicting nomophobia, highlighting their potential to deliver timely, data-driven insights for educators and mental health professionals. Such frameworks can facilitate early intervention, reduce anxiety, and enhance academic outcomes among university students.