In wire-TIG welding, the weld pool geometry depends directly on the heat input of the process. Many process parameters are impacting this heat input, a variation of the welding speed for example will have a strong impact on the weld pool penetration and consequently leading to a lack of fusion defect. For this reason, the paper focuses on this process and proposes a welding classification and prediction model to control the physics of the weld pool. More specifically, a neural network approach is applied on several experimental data to predict the class of the different welding configurations. Furthermore, a computer vision algorithm was used first to extract the weld pool contour and features using camera acquisition and image processing to set up an organized and listed database for each welding configuration and for each process parameters, namely current intensity, arc voltage, wire-feed rate, and travel speed. To validate the effectiveness of the proposed approach, a test split validation strategy was applied to train and validate the neural models. Several neural network layers with different size have been tested to obtain an accurate classification model with low errors and good performance scores. The results show a particular evolutionary trend and confirm that the process parameters have a direct influence on the weld pool, and so, too, on the final bead geometry. Finally, this study indicates that the neural network approach can efficiently be used to predict and classify welding process parameters for each configuration and could help later in designing a proper controller and a new real-time technic for welding quality.