Extrusion-based additive manufacturing (AM) technology has the potential for cost-effective 3D printing of intricate parts in the realm of AM. However, the persistence of stringing defects poses a significant obstacle, hindering the industrial utilization of 3D-printed components. This paper presents an investigation into stringing defects in 3D-printed PLA parts using the MobileNetV2 convolutional neural network (CNN) model and proposes an optimization methodology using the Taguchi design of experiments. The research involved two sets of nine experiments each for printing tensile specimens with PLA material on a Delta 3D printer and capturing images using a Raspberry Pi HQ camera. This ultimately led to the creation of a dataset of 1309 images for analysis. By optimizing printing process parameters like extruder temperature, printing speed, retraction distance, and retraction speed, the occurrence of stringing defects was effectively minimized. The MobileNetV2 CNN model trained using the generated dataset demonstrated an impressive defect identification accuracy of 98%. This integrated approach showcases a promising in situ defect detection technique that allows for real-time monitoring and optimization of the 3D printing process, leading to improved part quality and reduced waste in AM applications.