Huber Loss in Machine Learning: A Robust Alternative to MSE