Integrating Random Forest and Multifractal Filtering: A Novel Approach for Geochemical Data Mining of Oil and Gas in the Taiwan Strait Basin

This study focuses on the geochemical data mining of oil and gas in the Jiulongjiang Depression of the Taiwan Strait Basin, introducing and validating an innovative strategy that combines Random Forest (RF) modeling with multifractal filtering. Initially, the RF model was employed to analyze 24 geochemical indicators of oil and gas in the Jiulongjiang Depression. The results demonstrated that the model accurately delineates anomalous regions, particularly identifying an anomaly near the oil and gas center in the northwestern part of the depression. This area is characterized by favorable Eocene lacustrine source rocks and several oil and gas-bearing faults, which are conducive to hydrocarbon generation and accumulation. Subsequently, multifractal filtering was applied to the outputs of the RF model, effectively separating anomalies from background signals and highlighting the anomalous region in the northern part of the Jiulongjiang Depression, while also identifying a high-background area to the west of the depression. Furthermore, after down sampling the sample data, the RF model and multifractal filtering were reapplied, yielding results that were consistent with those obtained from the original dataset. This finding indicates that the combined strategy exhibits strong robustness against variations in data density, enabling stable identification of anomalous regions under different data density conditions. This achievement not only enhances the accuracy and reliability of geochemical data mining for oil and gas in the Jiulongjiang Depression of the Taiwan Strait Basin but also provides a new, efficient, and precise method for future oil and gas exploration, with significant practical applications and broad prospects.