The study investigated the hyperspectral reflectance responses to changes in the moisture content of potato tubers in a time series generated during oven drying. 17 chemometric preprocessing methods were used to eliminate the impact of spectrum noise on the spectral feature curve. The CatBoost, LightGBM, XGBoost, and other algorithms were used to obtain the effective feature spectra for hyperspectral images. Water content prediction models were derived by using selected feature spectra and the results indicated that the combined model based on the Lasso and XGBoost algorithms had the greatest prediction ability with the highest R2 value of 0.8908.