The impact of feedback types on behavioral engagement: a machine learning analysis

Feedback is a critical component of the learning process, yet the consequences of performance-feedback granularity on adolescents remains unclear. This quasi-experimental study evaluated short-term differences in academic performance, psychological states, and behavioral engagement among 288 secondary students assigned to three conditions: Precise Ranking, Vague Grading, and No Feedback (control group). Data were triangulated using post-intervention academic performance, six achievement emotions, and computer-vision-derived behavioral engagement proxies. Adjusted academic performance was descriptively highest in the vague-grading condition compared with the precise-ranking and no-feedback conditions, although the overall condition effect was not statistically supported. At the individual level, anger and boredom were significantly associated with lower academic performance. Two descriptive patterns emerged in the temporal analysis of behavioral engagement: mean engagement increased after feedback in the precise-ranking condition, while the proportion of observations classified as constructive increased under vague grading. These findings suggest that feedback granularity may produce differentiated response profiles across learning dimensions. Because the feedback conditions were allocated to only six classes, the condition-level patterns should be considered preliminary. By integrating academic performance, achievement emotions, and temporally detailed behavioral indicators, this study demonstrates the value of evaluating classroom feedback as a multidimensional process.