Improving the prediction accuracy of protein structural class: Approached with alternating word frequency and normalized Lempel–Ziv complexity