An Approach to Predict Hot Methods using Support Vector Machines

Most dynamic optimizers use feedback-directed adaptive optimization techniques. These techniques are expensive because of the profiling overhead. Although the recent trend has been toward the application of machine learning heuristics in compiler optimization, its role in identification and prediction of hotspots has been ignored. This approach evaluates a support vector machine (SVM) based machine learning technique in which static program features have been used to develop a model to predict program hot spots. The result has shown that, when trained with just ten features, the model predicts hot methods with an appreciable 70.93% accuracy.