I was playing with different machine learning algorithms with some data of mine which are really messy. While other algorithms were not showing much promise , using neuralnet I've achieved some form of consistency, about 76-78% correct prediction with random training set.
Actually I don't have much understanding about backpropagation, activation function or other stuffs. My results were obtained by experimenting with different parameters.
Now, how do I know if this performance of my model is improvable or not? Where to look to get some idea about what could be my next step?
This is my code for building the model:
neuralnet(result ~ factor1+factor2+...+factor13, ,data,algorithm='rprop+',hidden=15,rep=15, act.fct="logistic", err.fct = "sse", stepmax = 1e+06,threshold=0.01,linear.output = FALSE, likelihood = FALSE)
If the data are images, try cnn
Also, try dropout to avoid overfitting, and different network configurations (even ensembles if training time is small)
try to increase the number of neurones in hidden layers or the number of hidden layers to acheive the best performance.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training