Can k-NN imputation improve the performance of C4.5 with small software project data sets? A comparative evaluation