Low-permeability sandstone reservoirs containing low resistivity pays (LRPs) are widely encountered within the Triassic Yanchang Formation in the Jiyuan area, Ordos Basin, China. The resistivity of these reservoirs can be as low as 2·Ω m, leading to misinterpretation by conventional logging methods. Previous studies have primarily focused on resultant factors such as conductive minerals and clay minerals. However, how different diagenetic processes couple and control low resistivity remains unclear. This study focuses on the Chang 2 interval of the Yanchang Formation, integrating petrology, hydrochemistry, and machine learning (ML) to address these challenges. Results reveal that compaction-induced concentration of formation water and feldspar dissolution are critical controls on LRP formation. Dissolution processes not only release ions such as Na⁺ and Cl⁻, increasing formation water salinity, but also produce authigenic clay minerals like kaolinite, which obstruct pore throats and trap highly saline irreducible water, significantly reducing reservoir resistivity. Additionally, dissolution generates larger pores (diameter >10 μm), enabling more stable conductive networks and further lowering resistivity. In terms of fluid identification, the two newly constructed fluid-sensitive parameters exhibit strong explanatory capability, and the RF-PFN model outperforms other ML methods in distinguishing fluid properties. Prediction case studies show that LRPs preferentially develop in areas characterized by higher sandstone content, high porosity, and intensive dissolution effects. These findings provide critical insights and practical frameworks for evaluating and predicting similar low-permeability sandstone reservoirs worldwide.