Reliable multiclass cancer classification of microarray gene expression profiles using an improved wavelet neural network

► Studies were made on the proper initialization of wavelet neural networks (WNNs). ► A novel clustering algorithm was formulated and integrated in WNNs initialization. ► The improved WNNs were applied in heterogeneous cancer classification. ► Accuracies within 92-100% were obtained for four different microarray datasets. ► Comparison with general classifiers validates its beneficial potential.