Help - My Neural Network Does Not Work!

There are many ways in which an artificial neural network (ANN) can break down and not perform well. In this blog, we go through some of the essential requirements for getting a feed-forward ANN to work properly for a typical classification problem. The dataset chosen was the “Large Soybean Database” dataset from the UCI Machine Learning repository [Michalski & Chilausky, 1980] There are 15 classes and 35 categorical attributes in this dataset, some nominal and some ordered. We make the assumption that the data can be directly input into an ANN without creating k-1 dummy indicators (0,1) for each level of the factors, which would result in more than 100 binary features. (this could be done, but was not for the sake of efficiency). The number of objects was n=266 (there were n=307 original objects but we dropped records with missing data). <a href="https://medium.com/@peterson.leif.e/help-my-neural-network-does-not-work-fc1db171fb41">Read More</a>