Constructing Effective Machine Learning Models for the Sciences: A Multidisciplinary Perspective