Lung cancer classification using deep learned features on low population dataset

Classification of lung cancer using a low population, high dimensional dataset is challenging due to insufficient samples to learn an accurate mapping among features and class labels. Current literature usually handles this task through hand-crafted feature creation and selection. In recent years, deep learning is found to be able to identify the underlying structure of data through the use of autoencoders and other techniques. In this work, a deep autoencoder classification mechanism is proposed which first learns deep features and then trains an artificial neural network with these learned features. Experimental results show the deep learned classifier outperforms all other classifiers when trained with all attributes and same training samples. It is also demonstrated that the performance improvement is statistically significant.