Wavelet-based signal processing for evaluating and forecasting e-resource usage

This study evaluates and forecasts the usage of the OECD (Organisation for Economic Co-operation and Development) iLibrary database, subscribed to by the Library of the Grand National Assembly of Türkiye (GNAT). To capture dynamic and scale-dependent relationships between usage and external shocks we applied Morlet and Mexican hat wavelet coherence analysis, which are effective tools for continuous wavelet coherence analysis. The results of Morlet wavelet coherence revealed short- and medium-term co-movements driven by external factors, whereas Mexican hat coherence identified more persistent long-term relationships. Subsequently, the regression algorithms were applied on usage data decomposed by maximal overlap discrete wavelet transform (MODWT) and multiresolution analysis (MRA), providing a flexible framework for representing usage data at multiple resolution levels, while preserving temporal alignment and minimizing information loss. The findings revealed that usage patterns exhibited a multi-year synchronization with exogenous factors and regularized linear approaches such as ElasticNet, Lasso, and Ridge regression outperformed non-linear methods. SHapley Additive exPlanations-based interpretation, which quantified the contribution of each predictor of model outputs, showed that short-term fluctuations in usage data were the strongest predictors of future usage. The study demonstrates that e-resource usage should be understood and managed as a multi-scale, dynamically evolving process, and provides a framework that enables libraries to anticipate demand, align collections with institutional cycles, and respond to both short- and long-term changes.