With the proliferation of intelligent algorithms and online media ecosystems, college students are increasingly exposed to complex digital environments that influence their behavioral and psychological profiles. This study proposes an optimization model for student assessment systems in educational management, grounded in data fusion algorithms and structural equation modeling (SEM). Drawing upon a multi-dimensional theoretical framework, the study investigates how family experience, school environment, stress factors, and intelligent algorithmic recommendation contribute to students’ exposure to negative media content and, subsequently, to online behavioral misconduct. Data were collected from 372 college students using a validated Likert-scale questionnaire covering six latent variables and 18 measurement items. SEM results demonstrate that negative media content significantly mediates the impact of school experience, stressors, and algorithmic influence on behavioral misconduct, while family experience shows no statistically significant direct effect. Furthermore, a supplementary machine learning analysis using a Random Forest classifier revealed an F1-score of 0.83 and AUC of 0.89, highlighting the predictive power of fused variables such as algorithmic feedback and psychological stress in identifying students at risk.