Analysis of the Relationship between College Students' Mental Health Fluctuations and The Effect of the Civics Program Based on the ARFIMA Model

Abstract

This study establishes a methodological framework for quantifying the dynamic relationship between fluctuations in college students' mental health and the effects of ideological and political education courses based on the Autoregressive Fractional Integrated Moving Average (ARFIMA) model. By integrating multi-source heterogeneous data and combining time series feature representation techniques (Fourier transform, piecewise aggregation approximation), the study systematically analyzes the long-term memory characteristics of mental health. Experiments on public datasets (StudentLife, WeiBo) demonstrate that the ARFIMA model significantly outperforms baseline models in mental health level prediction, achieving a weighted average F1 score of 50.56% (an improvement of 7.27% over the best baseline SOTA), with a maximum F1 score of 97.05% on the WeiBo dataset. To address the issue of overfitting in small samples, a low-rank adaptive (LoRA) optimization mechanism was introduced, reducing the validation set loss from 1.051 to 0.956 and enhancing the model's generalization capability. After LoRA optimization, the emotion recognition model demonstrated high discriminative power for positive emotions (97.77% accuracy) and sadness (91.79%), but neutral emotion recognition exhibited confusion (90.12% accuracy). Related analyses of the effectiveness of ideological and political education courses showed that course learning outcomes were most strongly correlated with positive psychological qualities (r = 0.494), particularly in the dimensions of cognition (r = 0.481) and self-control (r = 0.479). Regression analysis further indicates that course content (β = 0.136, p < 0.05) and diverse development (β = 0.089, p < 0.05) are core predictive factors for cognitive qualities. This study provides data-driven decision support for optimizing ideological and political education courses and confirms the effectiveness of time series modeling in mental health interventions.