<p>Despite the extensive utilization of additive manufacturing technology and markedly fused filament fabrication across several industries, challenges such as dimensional and geometric inconsistencies persist, hindering its full-scale manufacturing capabilities. Therefore, it is imperative to study the impact of different process parameters on dimensional and geometric deviations of FFF printed parts and develop models that could correlate the process parameters with the output responses. This study investigates the effect of four critical process parameters, including layer thickness, printing speed, extrusion temperature, and extrusion width, on the dimensional and geometrical deviations of polycarbonate samples developed by the FFF process. Three regression-based machine learning (ML) models, including linear regression model (<i>LReg</i>), random forest regression model (<i>RFReg</i>), and extreme gradient boosting regression model (<i>XGBReg</i>), were trained and tested for predicting the dimensional and geometric deviations of FFF printed samples. The findings revealed that lower layer thickness, printing speed, and extrusion width, in combination with moderate values of extrusion temperature, provide the highest dimensional and geometric accuracy in FFF parts. It was also found that out of the three ML models, <i>XGBReg</i> models provided the highest accuracy and performance in terms of error metrics. <i>XGBReg</i> exhibited the lowest MSE, RMSE, and MAE values as 0.039, 0.199, and 0.152, respectively, with the highest value of R² as 0.9486 in cases of linear deviation measurements. Similarly, for diametral deviation, flatness error, and cylindricity error, <i>XGBReg</i> exhibited the lowest values of MSE, RMSE, and MAE and the highest R² values, indicating the highest prediction performance and accuracy of <i>XGBReg</i> models among all the three ML models studied.</p> Graphical abstract <p></p>