<p>In the oil and gas industry, assessment of porosity is critical for reservoir evaluation and management, serving as a fundamental aspect of reservoir research. However, the heterogeneity of sandstone reservoirs, such as those found in the Tahe oilfield, poses considerable challenges in accurately predicting porosity and permeability, especially in terms of declining production phases. In this paper, a novel hybrid machine learning (ML) model designed to improve porosity prediction using well log datasets from three wells in the Tahe oilfield is introduced. Datasets from wells S95 and T912 were used for training, while those from well S100 were used for testing. The novelty of this approach lies in the integration of four distinct ML models—random forest regression, gradient boosting regression, K-nearest neighbor regression, and multilayer perceptron regression—into a single-ensemble framework. This hybrid model leverages the strengths of each algorithm to reduce bias and variance and mitigate overfitting, thereby improving its generalizability across different well log datasets. Unlike previous methodologies, this ensemble technique directly addresses the complexities of porosity prediction in heterogeneous reservoirs. The hybrid model demonstrates superior predictive accuracy, outperforming individual models and closely matching actual porosity values. This model achieved the highest coefficient of determination (94.6%) and the lowest root mean square error (1.899), emphasizing its strong generalizability to unseen data. These results highlight the effectiveness of the proposed hybrid model in accurately estimating porosity in Triassic reservoirs, providing an innovative alternative to traditional measurement techniques. Overall, this approach provides remarkable contributions to the oil and gas engineering field, enhancing reservoir evaluation and management.</p>