Surface roughness is a key parameter in machining processes that directly affects product quality and performance. This study investigates the use of machine learning models, specifically neural networks (NN) and recurrent neural networks (RNN), to predict surface roughness based on sensor data collected during machining operations. The dataset consists of vibration, acoustic and current signals that were pre-processed and used as input features for the predictive modelling. The results show that both models effectively capture the relationship between machining parameters and surface roughness. The RNN model outperforms the NN in most accuracy metrics, achieving lower mean squared error (MSE), root mean squared error (RMSE) and mean squared logarithmic error (MSLE), indicating better generalisation. Specifically, the RNN achieved an MSE of 0.0021, demonstrating its superior predictive ability. However, the NN lower mean absolute error (MAE), mean absolute percentage error (MAPE) and median absolute error (MedAE), indicating more stable absolute error performance. This study highlights the potential of machine learning in optimising machining processes, providing a data-driven approach to improve surface quality prediction, reduce material waste and increase industrial efficiency.