To address the challenges of obtaining starch and moisture content information from fermented grains during the fermentation of light-flavor baijiu and the limitations of traditional point prediction models, an interval prediction model was proposed for measuring starch and moisture content in fermented grains during fermentation. Firstly, input parameters for the starch and moisture prediction models were determined through a comprehensive analysis of easily measurable potential explanatory variables, such as physicochemical parameters of Huangshui and fermented grain temperature, using correlation coefficients and least-angle regression. Secondly, a high-accuracy point prediction model for starch and moisture content in fermented grains was established by optimizing hidden layer weights and thresholds of the extreme learning machine using a genetic algorithm. Finally, four different Bootstrap methods were employed to estimate point prediction results, systematic errors, and random errors, thereby constructing prediction intervals for starch and moisture content in fermented grains at different confidence levels. Results indicated that, at a 99% confidence level, the interval prediction model based on the Residual Bootstrap method exhibited superior performance. On the test set for starch and moisture content in fermented grains, the coefficients of determination for points predictions were 0.998 9 and 0.109 1, respectively, and the root-mean-square error for point predictions were 0.920 3 and 0.802 1, respectively. The prediction interval coverage probabilities were 100% and 0.728 0%, and the mean prediction interval widths were 100% and 4.339 9%, respectively. This study provides reliable predictions for starch and moisture content in fermented grains during the production of light-flavor baijiu, offering valuable references for production guidance and improving transparency in the fermentation process.