Classification can be used to predict unknown functions of proteins by using known function information. In some cases, multiple sets of data are available for classification where prediction is only part of the problem, and knowing the most reliable source for prediction is also relevant. Our goal is to develop classification techniques to find the most predictive of the multiple data sets that we have in this project. We use existing classification techniques like linear and quadratic classifications and statistical relevance measures like posterior and log p analysis in our proposed algorithm, which is able to find the data set that is expected to give the best prediction. The proposed algorithm is used on experimental readings during cell cycle of yeast and it predicts the genes that participate in cell-cycle regulation and the type of experiment that provides evidence of cell cycle involvement for any particular gene.