Software code is composed of several components (e.g., several Java classes). Testing all these components can be a very expensive task. If we know which components are likely to be defective, we can concentrate testing on these components, increasing the chances of finding software defects while reducing testing effort. The task of software defect prediction is concerned with predicting which software components are likely to be defective, helping to increase testing cost-effectiveness. In this talk, I will show how software defect prediction can be performed by using automated machine learning approaches. I will also go through some important issues to be considered when using such automated approaches.