An Explainable Wavelet-Based Feature Decomposition and Machine Learning Framework for Land Cover Classification

Accurate land cover classification is essential for environmental monitoring, urban planning, and resource management. Conventional classifiers trained on raw spectral bands are often limited by noise, inter-class spectral similarity, and intra-class variability. This study introduces a wavelet-based feature decomposition and machine learning framework to address these challenges. Landsat-8 Operational Land Imager (OLI) Level-2 surface reflectance data were pre-processed and decomposed using a one-dimensional discrete wavelet transform to isolate low- and high-frequency components. The decomposed features were concatenated with raw bands to form an enriched dataset, which was used to train and validate three supervised classifiers: Random Forest (RF), Gradient Boosting (GBM), and Decision Tree (DT). Model performance was evaluated using 5-fold cross-validation, and the best-performing model, the Random Forest Wavelet Transform (RF_WT), achieved the highest macro-F1 score (0.865). Explainable AI (Gini importance, permutation importance, SHAP (SHapley Additive exPlanations) values) confirmed the complementary role of raw and decomposed features, with wavelet-derived detail components in the Short-Wave Infrared (SWIR) and visible bands strongly influencing classification. The RF_WT model, when compared to RF, DT and GBM classifiers trained on raw spectral bands across all classes, achieved the highest macro-averaged accuracy values (UA = 0.876, PA = 0.864, F1 = 0.865), compared with RF (UA = 0.864, PA = 0.852, F1 = 0.853), GBM (UA = 0.849, PA = 0.832, F1 = 0.832), and DT (UA = 0.815, PA = 0.808, F1 = 0.809). The approach is valuable in support of ecological monitoring and resource management. However, challenges in settlement classification highlight the limitations of medium-resolution data. Future work could integrate higher-resolution imagery or multi-temporal data for improved performance.