Apple pomace is an agro-industrial byproduct high in dietary fiber, with composition varying based on the extraction process. In this study, components of insoluble fiber (cellulose, hemicellulose, lignin, NDF, and ADF) were quantified using the Van Soest method in pomace flours obtained from two industrial processes: turbo extraction (Process A) and mechanical pressing (Process B). The data served to calibrate and validate predictive models using near-infrared (NIR) spectroscopy. Chemometric tools (PCA, PLS-DA) effectively discriminated between the two types, with sensitivity, specificity, and accuracy reaching 1, 1, and 100%, respectively. Artificial neural network (ANN) models accurately predicted insoluble fiber contents, with R2 values up to 0.998 in calibration and 0.989 in validation. These results demonstrate that NIR combined with chemometric and machine learning techniques is a rapid, non-destructive, and effective method for characterizing the chemical composition and insoluble fiber fractions in apple pomace from different industrial sources.