Hybrid Deep Learning Using LSTM and K-Means for Temporal-Spatial Library Demand Prediction

View at publisher Summarise with VIRAT
Authors: Sasalak Tongkaw, Seree Chanah, Aumnat Tongkaw
Cited by: 0
References: 17

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Academic libraries require sophisticated predictive models to optimize resource allocation and anticipate user demand patterns. This study presents a novel hybrid approach combining K-means clustering with LSTM networks to predict library borrowing patterns through temporal-spatial user behavior analysis. Using 26,857 borrowing records from Songkhla Rajabhat University, we compare the performance of static clustering versus dynamic deep learning prediction. Kmeans clustering analysis identified optimal segmentation with$\mathrm{K}=2$clusters, achieving a silhouette score of 0.280 and CalinskiHarabasz index of 220.96. The clustering revealed two distinct user segments: high-activity users (156 users, 6.38 %) averaging 77.17 books with July peak activity, and low-activity users (2,291 users, 93.62 %) averaging 5.99 books with March concentration. Feature importance analysis showed total books borrowed (35.59) and unique books (21.43) as primary clustering determinants. The hybrid LSTM-enhanced approach significantly outperformed traditional clustering by incorporating temporal dependencies. LSTM models achieved MAE of 0.143 and RMSE of 0.187 for demand forecasting, demonstrating superior predictive capability compared to static clustering approaches. The integration of cluster-specific LSTM architectures enabled accurate$\mathbf{1 2}$-month demand predictions while maintaining user segment interpretability. Comparative analysis reveals that while K-means provides essential user segmentation foundation, LSTM networks add crucial temporal forecasting capabilities. The hybrid methodology achieved 23% improvement in prediction accuracy over baseline clustering methods, enabling proactive resource management. This approach transforms static user profiling into dynamic demand prediction, offering libraries evidence-based tools for strategic planning and operational optimization in academic environments.

Fields of Study & Subjects

Data Mining and Machine Learning Applications Forecasting Techniques and Applications Web and Library Services

Publication details

Journal / source
International Conference on Cognitive Systems and Computer Interaction (ICoSCI)
Pages
1–6
Publication date
15 January 2026
Language
EN