Hello Chau,

I believe it is important to know whether you will be using direct encoding or indirect encoding for your neural network. Please find below a few links to my papers which could be useful.

Conference Paper Enhancing the performance of neural network classifier using...

Article Application of Predictive Coding in Neuroevolution

Any encoded input, either real or integer values will do. If the input values are huge in size, it is better to normalize the same with any standard procedure. With normalized values, there won't be any overflow across the layers.

Dear Anh

I work with NN more than 20 years. Sorry, I do not have comparison with other intelligent systems. NN can swallow any type of data. Different coding can help you to improve result, also input data normalisation can help. With Border pairs method you could eliminate useless input patterns. Sometimes are more than 90% of all patterns useless. In such a cases effect of elimination is very powerfull.

Conference Paper Border Pairs Method – Constructive MLP Learning Classificati...

Chapter Advances in Machine Learning Research

Hi Dr. Ploj, thank you very much for your answer. Can you explain me some basic ideas on how the neural network differs from other predictive models such as regression or support vector machine. Thank you

The most popular approach is normailizing the input data into a proper range, such as [-1, 1], [0, 1], and so on. Normalization algorithms are also abundant for different data requirements, like sigmod function for [0, 1], tanh function for [-1, 1], etc..

It can be helpful.

Regards

Hubert

Article The Influence of Input Data Standardization Method on Predic...