The software system has a huge amount of programming codes, several procedures, and modules. Due to this, it is too complex to understand. For the better experience of the user, software must be running efficiently. If any module gets any type of defect, it may harm the field of health care, defense, education, and so on. It can also be expensive in terms of money, effort, and reputation for a company. It is very tedious work to identify the defects in the module. To allot resources efficiently, reduce costs, and enhance the performance of software, the prediction of defective software is very necessary and it may help developers to save time and reduce the development cost. In the testing phase of software, we get most of the defects which reduces the quality of the software. According to the report (Herb Krasner), the testing phase takes more time as compared to other phases and takes more than 50 percent cost of the software where finding and fixing the defect takes place. This paper uses five datasets “MC1, JM1, KC1, CM1, and PC1” of the NASA repository for analysis. This repository contains 13 datasets with different instances from range 127 to 17001. Firstly, we use a Support Vector Machine (SVM), Random Forest (RF), and two Na¨ıve Bays (NB) algorithms namely Gaussian Naive Bayes (GNB) and Bernoulli Naive Bayes (BNB) to calculate the results for each dataset. Secondly, we use PCA for dimensionality reduction and calculate the results before and after applying PCA to all aforementioned algorithms. Finally, we observe that after using PCA, the results of all the algorithms have been improved.