Forecasting nuanced shifts in student engagement from longitudinal experiential (LE) data—multi-modal, qualitative trajectories of academic experiences over time—remains challenging due to high dimensionality and missingness. We propose a natural language processing (NLP)-driven framework using large language models (LLMs) to forecast binary engagement levels across four dimensions: Lecture Engagement Disposition, Academic Self-Efficacy, Performance Self-Evaluation, and Academic Identity and Value Perception. Evaluated on 960 trajectories from 96 first-year STEM students, our three-tier approach—LLM-informed imputation to generate textual descriptors for missing-not-at-random (MNAR) patterns, zero-shot feature selection via ensemble voting, and fine-tuned LLMs—processes textual non-cognitive responses. LLMs substantially outperform numeric baselines (e.g., Random Forest, LSTM) by capturing contextual nuances in student responses. Encoder-only LLMs surpass decoder-only variants, highlighting architectural strengths for sparse, qualitative LE data. Our framework advances NLP solutions for modeling student engagement from complex LE data, excelling where traditional methods struggle.

Longitudinal experiential data offers rich insights into dynamic human states, yet building models that generalize across diverse contexts remains challenging. We propose ConText-LE, a framework that systematically investigates text representation strategies and output formulations to maximize large language model cross-distribution generalization for behavioral forecasting. Our novel Meta-Narrative representation synthesizes complex temporal patterns into semantically rich narratives, while Prospective Narrative Generation reframes prediction as a generative task aligned with LLMs’ contextual understanding capabilities. Through comprehensive experiments on three diverse longitudinal datasets addressing the underexplored challenge of cross-distribution generalization in mental health and educational forecasting, we show that combining Meta-Narrative input with Prospective Narrative Generation significantly outperforms existing approaches. Our method achieves up to 12.28% improvement in out-of-distribution accuracy and up to 11.99% improvement in F1 scores over binary classification methods. Bidirectional evaluation and architectural ablation studies confirm the robustness of our approach, establishing ConText-LE as an effective framework for reliable behavioral forecasting across temporal and contextual shifts.