Surface Roughness Prediction in Additive Manufacturing: Presenting the Power of Neural Networks Compared to Linear Regression

This paper investigates surface roughness prediction in additive manufacturing through a comprehensive comparative analysis of linear regression (LR) and neural network (NN) models. Employing [Formula: see text]-means clustering, we identify four distinct clusters within the experimental data, each closely associated with specific 3D printing parameters. Within each cluster, we explore the optimal combination of factors that contribute to surface roughness and power efficiency. The main objective focuses on predicting a target variable, with an emphasis on evaluating model performance via key metrics such as [Formula: see text]-squared ([Formula: see text] 2 ), adjusted [Formula: see text]-squared, predicted [Formula: see text]-squared, mean squared error (MSE), and correlation coefficient ([Formula: see text]). Our study’s results illuminate the robust predictive capabilities of both LR and NN models. However, it becomes evident that the Neural Network model outperforms Linear Regression. It exhibits excellent performance metrics, characterized by higher [Formula: see text] 2 and correlation values, reduced MSE, and greater resilience to outliers. This pronounced disparity underscores the Neural Network model’s exceptional suitability for tasks requiring precise predictions and the identification of nonlinear patterns, particularly in the field of surface roughness prediction in additive manufacturing. These findings emphasize the key role of advanced machine learning techniques, illustrated by neural networks, in achieving precision within similar domains.