Machine Learning approaches are helpful & have well-tried to be helpful in resolution issues & technical problems that lack data. In most cases, the package domain issues may be characterized as a method of learning that depends on the assorted circumstances and changes of the technical issue being addressed in keeping with the principles of machine learning, a prophetic model is made by exploitation machine learning approaches and classified into defective and non-defective modules. Machine learning techniques facilitate developers to retrieve helpful data when the classification of kinds of technical problems being addressed in an exceedingly specific field. This successively permits them to analyze knowledge from totally different views, which may be used because of the formation base of constructive concepts & varied techniques to handle the technical problems. Machine learning techniques are well-tried to be helpful within the detection of package bugs. during this analysis prediction by Convolution based mostly feature choice and Learning by Random forest. In the proposed approach, the accuracy and precision always improve and it also improves class wise. There is a significant enhancement in defective and non-defective class prediction as the random forest non-linearity features help to improve the selection of effective parameters by bagging approach. In the proposed approach, hybridization of three approaches such as deep learning, machine learning and sampling approach is done which significantly improve overlapping of features and imbalance of class like KC2 dataset.