Optimizing olive oil extraction process could be a key aspect to improve the quality of the final product. This study aims to develop an automatic olive oil and water content percentage predictor within olive paste or pomace samples using near-infrared (NIR) spectroscopy and machine learning. The samples were collected from two olive mills in Tuscany, Italy, and their NIR spectra were acquired. To ensure more accurate results a wide range for the olive oil content (3.23% to 30.54%) and water content (43.23% to 75.54%) has been collected. Several techniques are considered, trained and evaluated using k-fold cross-validation. The results shows that the Bayesian Ridge Regression model outperforms the other techniques, achieving a Root Mean Squared Error (RMSE) of 3.22% and 2.03% respectively for olive oil and water content considering the original spectral data. After denoising the spectral data, the model's performance improves further, reaching an RMSE of 3.00% for olive oil and 1.85% for water on the hold-out test set. Starting from these findings it is possible to understand the effectiveness of the developed of automatic predictor to accurately estimate olive oil and water percentages within olive pomace. Introducing this technique in a real scenario could help to optimize the olive oil extraction process and the utilization of this agricultural by-product, also enabling the possibility to rapidly and objectively understand the quality of the product.