Constructing New and Better Evaluation Measures for Machine Learning