Machine Learning-Based Prediction of Surface Roughness in Fused Deposition Modeling Using Multi-Parameter Process Data

Surface roughness is a key indicator of printed‐part quality in fused‐deposition modeling (FDM) and is strongly influenced by geometric and thermal process parameters. This study investigates the predictive behavior of surface roughness using a multi-parameter dataset consisting of layer height, wall thickness, infill density, infill pattern, nozzle temperature, bed temperature, print speed, material type, and fan speed. Categorical parameters were label-encoded and fed into four machine-learning models: Generalized Linear Model (GLM), Gradient Boosting Machine (GBM), Deep Learning (MLP), and a Stacked Ensemble combining GBM and MLP with GLM as the metalearner. All models were trained directly on raw data without normalization to preserve parameter scale effects. Layer height and nozzle temperature emerged as the most influential parameters, contributing over 85% of total predictive importance. Model comparison showed that GLM achieved the highest accuracy (R² = 0.789), producing stable predictions with well-behaved residuals, while nonlinear models underperformed on unscaled features. Diagnostic results confirmed that surface roughness is governed mainly by geometric resolution and melt-flow behavior. The findings provide a robust, data-driven understanding of roughness formation and offer a reliable prediction framework for optimizing FDM print settings.