We target the problem of software defect prediction. The main aim is to build a strong classifier that is capable of detecting software defects. Imbalance of data is a major problem associated with software defect datasets, which can lead to classification problems such as over-fitting and high false positive rates. To mitigate these problem, we apply an oversampling technique named Synthetic Minority Oversampling technique (SMOTE). SMOTE rebalances the original dataset by introducing synthetic instances to minority class. The newly balanced datasets are then fed to a variety of tree-based classifiers where Optimized Forest algorithm provides a superior performance. As of the relatively high dimensionality of the feature space, of sixty-one features, we apply BestFirst feature selection algorithm and experiment the classification performance of Optimized Forest in the reduced space. Overall, our results indicate that SMOTE is crucial for solving the imbalance of the dataset. SMOTE, in conjunction, with Optimized Forest classification algorithm is effective in predicting defective modules. However, feature selection is not very efficient for the problem in stake.