In this study, we introduce a pioneering approach that leverages advanced machine learning and ultrahigh-resolution Fourier transform ion cyclotron mass spectrometry (FT-ICR MS) data to predict the distribution of the total acid number (TAN) in true boiling point (TBP) distillation cuts from crude oil. By employing partial least-squares (PLS) regression and ordered predictor selection (OPS), we achieved robust predictive models with high accuracy, evidenced by low root-mean-square error of calibration (RMSEC) and strong correlation coefficients (Rc). Our analysis of 36 diverse crude oil samples revealed significant variations in chemical composition, with nitrogen- and oxygen-containing compounds playing key roles in influencing TAN values. Through the use of volcano plots, we identified critical molecular classes that drive changes in TAN. The predictive models demonstrated remarkable consistency between predicted and actual TAN values, particularly in samples with a higher TAN, further validating their reliability. Significantly, our method overcomes the limitations of traditional ASTM testing by requiring smaller sample volumes while still providing accurate TAN predictions. This novel approach offers a powerful new tool for the molecular characterization and behavioral forecasting of complex mixtures, enabling a more efficient pathway for sample analysis when resources are limited.