Object-oriented (OO) code has many dependencies among the classes and different types of changes that often have an impact during the maintenance of the software. In this paper, we proposed a technique for change impact analysis (CIA) with fault prediction (FP) using machine learning techniques for OO software. In proposed method an intermediate OO program representation is proposed using a graph which detects the difference between the original program and the modified program. For fault prediction, class level metrics are extracted from KC1 data set and 14 machine learning algorithms are trained on dataset. The genetic algorithm (GA) and correlation are used to extract significant features from the dataset. Trained prediction models are evaluated using classification accuracy (Acc). Our proposed approach consider various types of changes possible in the program, and test cases are selected to test the modified code during regression testing and finds the fault in classes. Among the 14 machine learning algorithms random forest (RF) giving best accuracy than other algorithms.