Hello All,
I am using a medical dataset with 903 records and 120 numeric attributes (mostly categorial and some continuous).
I have tried multiple clustering and classification algorithms on it.
Now I want to explore neural networks for the same, if possible to get fuzzy classification values.
Kindly suggest how to proceed towards this and which neural network variant should I go for .
Actually one of your essential works is to choose right encoding method.
Then you can utilize tons of NNs to check which one works for you. By searching "neuro-fuzzy" in github/google, you will find loads of codes helping you to complete your research!
I would recommend this free ebook by Michael Nielson for getting started: http://neuralnetworksanddeeplearning.com/
Cheers, Raoul
For numerical, you can use regression models, also data mining approaches, e.g., decision tree, SVM, genetic algorithms, etc. See also https://ieeexplore.ieee.org/abstract/document/227388
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.307.4185&rep=rep1&type=pdf
https://towardsdatascience.com/neural-network-algorithms-learn-how-to-train-ann-736dab9e6299
Most likely, you are solving the problem of automating diagnostics in medical practice.
I would advise you to pay attention to such issues as redundancy or insufficient completeness of the space of symptoms.
When it comes to diagnostics, decision rules generated by experts work well here. There are methods that can significantly reduce the dimension of the data presented to the expert for assessment. The result of the expert assessment applies to the entire area of diagnostics.
If this approach is of interest, then I know the works of O.I. Larichev. Pay attention to the section: Building Expert Knowledge Bases.
I have attached several books (in Russian).
Best regards, A. Nikiforov
There is a good example for Binary Classification Tutorial with the Keras Deep Learning Library
https://machinelearningmastery.com/binary-classification-tutorial-with-the-keras-deep-learning-library/
- Badges
- Science topic