Accurate prediction of hydrocarbon presence in subsurface formations remains an open problem in petroleum exploration, particularly in complicated geology and petrophysical datasets with imbalance. Conventional machine learning algorithms such as the Random Vector Functional Link (RVFL) network have shown promise but are prone to performance sensitivity due to human-based parameter adjustment and the inability to effectively address class imbalance. This study avoids these limitations by proposing an optimization approach based on swarm intelligence that integrates the RIME algorithm and RVFL for better predictability. This study introduces a novel approach utilizing swarm optimization techniques for determining the presence of oil in wells. The method is based on the principles of Swarm Intelligence (SI), a well-known machine learning methodology inspired by the collective behavior of decentralized, self-organized systems. This research particularly harnesses SI for analyzing geological data for the prediction of whether a well contains oil or not. The core of the swarm-based approach lies in its ability to efficiently process vast and complex datasets to identify patterns indicative of oil presence. This is achieved by estimating and predicting the key petrophysical parameters for hydrocarbon reservoirs (e.g., water saturation, total porosity, & shale volume). This swarm-optimized approach benefits from the space-searching capability of the swarm algorithms, leading to more accurate and speedy predictions. Random Vector Functional Link (RVFL) algorithm is used to optimize the selection of relevant geological features and parameters, enhancing the model’s predictive accuracy. RVFL processes well log data to predict gas-bearing intervals, while RIME automates parameter selection (e.g., hidden nodes, activation function), achieving up to 99.1% accuracy on imbalanced datasets. Applied to well log data from the Abu Madi Formation, the hybrid model optimizes key petrophysical parameters (e.g., water saturation, porosity, shale volume), offering superior accuracy, sensitivity, and computational efficiency compared to existing methods like AHA and GWO. This approach represents a significant advancement for hydrocarbon exploration workflows, enabling scalable gas prediction in complex geological settings.