Comments

Nice page!

1. In your experiment results, does "original dataset" mean no regularization?

2. Is accuracy a good metric for disease diagnosis? Should you weight more for false negatives?

YanZhao (talk)02:13, 21 April 2016

Hi Yan Zhao,

Thanks for your critique.

1. Original dataset means there is no feature reduction but I still use L1-regularization to train the model.

2. For this page, my objective is to reduce feature dimensionality to make the learning model simple and the prediction of the model more explainable to the doctors without significant loss of predictive accuracy. So accuracy is one metric, the other is feature dimensionality. If the model trained with new data sets still has good predictive accuracy after feature selection, I can assume selected features are important for the learning objective and those removed features are less relevant or irrelevant.

Sincerely,

Ke Dai

KeDai (talk)02:47, 21 April 2016