HOW TO LOWER MSE VALUES IN NARX?

I am having a hard time figuring out how to lower the MSE values for our model. We used NARX in our model, we have 1734x2 for the predictors and 1734x1 for the response. We are not satistified in our validation performace because of the high MSE values, we tried increasing the number of neurons or the number of delays, and we also trained the data multiple times. I attatched the photo for our model summary. Please help me.

Answer 1

[Lokesh]

Hi Vane,

I understand that you have not achieved satisfactory results in validation performance for your model because of high “Mean Squared Error (MSE).”

Here are some general suggestions that can help you reduce the “Mean Squared Error (MSE)” values for your “NARX” model:

• Increase training data: If possible, gather more training data to improve the model's ability to learn patterns and enhance its performance.
• Regularize the model: Apply regularization techniques such as L1 or L2 regularization to prevent overfitting and improve the model's generalization ability.
• Adjust the network architecture: Experiment with different network architectures by varying the number of neurons, layers, and delays.
• Normalize the data: Scale your input and output data to a similar range. Normalization prevents features with large values from dominating the training process and can improve the model's performance.
• Adjust training parameters: Try different learning rates, number of epochs, and batch sizes. Additionally, explore different optimization algorithms ('trainlm', 'trainbr', 'traingd') to see if they yield better results for your specific problem.
• Explore alternative models: If the NARX model is not providing satisfactory results, consider exploring other types of models that might be more suitable for your specific problem.

Please refer to the below mentioned documentations to know more about overfitting and regularization :

Overfitting: https://in.mathworks.com/discovery/overfitting.html

Regularization: https://in.mathworks.com/discovery/regularization.html

I hope you find this helpful.

Best Regards,

Lokesh