The rubidium/strontium (Rb/Sr) ratio in loess deposits is a widely recognized indicator of pedogenesis and East Asian Summer Monsoon intensity on the Chinese Loess Plateau (CLP). However, its significance in Central Asian loess remains underexplored. This study utilizes machine learning techniques, including Random Forest and Partial Least Squares Regression to investigate the factors influencing the Rb/Sr ratio in the Ili Basin, northeastern Central Asia (NCA). The results reveal a strong correlation between the Rb/Sr ratio and spring to early summer temperature, particularly in March, May, and June. Unlike in the CLP, where leaching primarily drives Rb/Sr variations, vegetation dynamics, which are influenced by temperatures from March to June, are the dominant influence in the Ili Basin. Based on these findings, a stacked Rb/Sr record was reconstructed, spanning the Last Glacial Period to the mid-Holocene. This record is closely similar to NCA temperature variations and related to glacial dynamics on orbital and millennial scales. The stacked Rb/Sr sequence indicates that spring to early summer temperatures in NCA were lowest during Marine Isotope Stage (MIS) 4 and MIS 3b, then increased to peak around 12.5 ka, and declined from the early to mid-Holocene. Furthermore, a significant warming trend from MIS 3a to MIS 2 is linked to changes in sea surface temperature in the northern tropical Indian Ocean, which influenced the Hadley cell. These findings position the Rb/Sr ratio as a novel paleotemperature proxy for loess in NCA, offering new insights into regional paleoenvironmental dynamics.