How to use ensemble learning to combine multiple trained pattern recognition neural networks?

I saved the neural networks (patternnet) with seperate classes

for example:

Neural Network 1:

3-class problem

Neural Network 2:

Another 3 class problem

Neural Network 3:

Another 3 class problem

Is there any way to use ensemble learning to combine these neural networks to make an ensemble of say 9 classes.

can we pass input to the (combined) ensemble to classify the class or predict any one class from the 9 classes.

I am new to ensemble learning..

Answer 1

[Varun Sai Alaparthi]

Hello Raja,

I don’t suggest ensembling these networks as they are trained for different classes. For example, Neural network 1 isn’t aware of the classes that Neural network 3 is trained on. So, it might affect the final performance of the ensembled model.

Ensembling would be effective if all the neural networks are trained for 9 classes.

The better alternative would be to retrain a single model for 9 classes or ensemble models trained for all 9 classes.

I hope this information helps and please reach out for any further issues.

[RAJA SEKHAR BATTU]

Hi @Varun Sai Alaparthi, thank you for the answer. I miscommunicated. I have all the neural network trained with same three classes. I am trying random Forrest method.Iam quite new in this ensemble learning. Can you suggest me a way to start with ensemble learning and also combine patternnet neural networks. I mean, is it possible to combine the trained neural networks or I need to start again to train.

My ultimate goal is to find a way to combine the multiple trained neural networks to form a group for classification.

[Varun Sai Alaparthi]

Hello Raja,

Yes is it possible to combine neural networks using ensembling.There are different ensembling methods that you can leverage to combine multiple neural networks like stacking ,boosting etc. Please check which technique works best for your use case.

Please refer to this links for information on different ensembling techniques :

• Ensemble Methods in Machine Learning: What are They and Why Use Them? | by Evan Lutins | Towards Data Science
• Ensembling for neural networks
• Ensemble Learning: Stacking, Blending & Voting | by Fernando López | Towards Data Science