To detect the defect occurred during arc welding process, various techniques are developed such as x-ray, sonic detection and etc. However, since these conventional defect detection techniques are rely on human judgment or heuristics, the defect detection process requires many resources such as time, equipment. To handle these problems, we developed A system which predict the welding defect using current, voltage and wire feed values. The system collects current, voltage and wire feed values from welding machines and determine the start and end time of welding process. Then, the system determines the defect is occurred or not during welding process using clustering based machine learning technique. After defected is predicted the system, also, calculates the probability of defect type using recurrent neural network. We expect the system can reduce resources which is required to detect welding defect by removing the requirement of human judgment.