Every company requires reliable and effective software to fulfill its requirement, at a low cost. Defective software can be dangerous in all fields whether you are using it in development fields, medical fields, production fields, or any other. Prediction of defects at an early stage can help us to save money and provide reliable software but for prediction, it is mandatory to test the software and all its module, and it will take more than 50% cost of the total cost of the software. ML is very famous at this time and provides different algorithms for analysis and prediction like SVM, DT, RF, etc. This paper is basically an analysis of all machine-learning techniques from recent years which may provide helps to the researchers in SDP because still there is no proper tool or technique which can predict and remove the defect of software.