Abstract: The accurate identification of water-flooded zones in low-porosity low-permeability glutenite reservoirs is particularly challenging due to their strong heterogeneity and highly ambiguous log responses. To enable accurate efficient classification of waterflooding levels, this paper proposes a hybrid model (SSA-KELM) that integrates the kernelized extreme learning machine (KELM) with the sparrow search algorithm (SSA). An analysis of log responses for various waterflooding levels identifies waterflooding-sensitive curves, which serve as the dataset for model training and testing. This is followed by using SSA to optimize the key hyperparameters of the KELM model. This paper presents a final comparative analysis between the proposed SSA-KELM model and five benchmark models: traditional extreme learning machine (ELM), particle swarm optimization-based ELM (PSO-ELM), KELM, genetic algorithm-optimized KELM (GA-KELM), and PSO-KELM. The experimental results indicate that an optimal sparrow population of 200 yields the best KELM model with the optimal hyperparameters of C = 97.39 and gamma = 0.3. This SSA-KELM model achieves the highest classification accuracy and best generalization performance. The model delivers an accuracy rate of 85.4% in a practical well-log application, outperforming all benchmark models. This study provides a novel effective approach for evaluating waterflooding levels in low-porosity low-permeability reservoirs.