Although some studies have yielded promising results, accurately predicting individual mobility remains a significant challenge due to the inherent complexity and variability of personal movement patterns. We study the hidden multi-level structure ‌consisting of‌ multi-granularity long- and short-term patterns in personalized mobility behaviors, we introduce the Multi-layer Spatio-Temporal LSTM-Transformer Model (MSTLTM), which integrates multi-layer structural encoding with spatio-temporal information within an LSTM-Transformer architecture. The LSTM-Transformer component is designed to effectively capture complex sequential dynamics and long-range dependencies. By leveraging a hierarchical, multi-level design, our model uncovers latent structural patterns within mobility sequences—an essential factor for enhancing prediction accuracy. Comprehensive experiments conducted on three widely used public datasets confirm that our approach consistently outperforms existing state-of-the-art methods.