I have looked on the internet and found a lot of discussion about image classification problems, but almost none of just classification problems without images. I am trying to build a model with Keras using Python, I have a dataset of 827 features (43 are integers, and the rest is one-hot encoded) and 24,800 samples. My target is 3 classes one-hot encoded as well ( [1 0 0] [0 1 0] [0 0 1] ).

I have found a lot of complex image recognition networks but only simple deep networks for non-image problems. I need help on how I can make the deep neural network architecture for my problem.

  • $\begingroup$ Welcome to DS.SE! May I ask why you need a complex network for your task? Image classification tasks are usually complex (in terms of the number of features and their complicated linkage) by nature, so we need complex networks (in terms of the number of parameters) to address them. Your dataset has 827 features which is as small as a set of RGB images of size 17x17. $\endgroup$ Commented Jun 14, 2019 at 10:44
  • $\begingroup$ If a large number of features is your problem, you can use AutoEncoders to reduce the number of features without losing their significance. $\endgroup$ Commented Jun 14, 2019 at 11:53

1 Answer 1


There are actually a lot of examples out there which show how to apply Keras to different types of data. A really good source is F. Chollet's book "Deep Learning with Python". The code is online. You could start with the tutorial on predicting house prices. It is a regression problem, but you can easily change it to a classification problem.

A general remark: Neural nets (such as Keras) usually work well with very complex data (images, sound, text). For "normal" data, other approaches are often better suited (e.g. boosted trees). However, with the large number of features you have, NN is worth a try.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.