I have used ANN, ANFIS, wavelet-ANN and wavelet-ANFIS for water quality prediction. I have used highly correlated variables as inputs (Pearson correlation>0.5). The R2 for ANN, ANFIS, best wavelet-ANN and best wavelet-ANFIS models were 0.9936, 0.9954, 0.996 and 0.9967, respectively. In my research the wavelet-ANN and wavelet-ANFIS showed minimal improvement in comparison with single ANN and ANFIS. Is the minimal improvement because of high correlation of inputs and output?
.
with a baseline R2 of 0.9936, there is not much room left for improvement !
such high R2 values are pretty uncommon when learning from real data ... it might be a hint of overfitting : are those values reported "out of training set" ? (that is on a sample of data not used for training the model)
.
All of the developed models are correct and they dont have any problem.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence