Hi,
I extracted the features of Left and right eeg signals and classified into left and right hand from BCI compettition III dataset IVa. . I got accuracy of 60%. In matlab, can you please guide me how to correct my errors /misclassification by deep learning methods(or) machine learning methods which helps to improve 100% accuracy...
I don't know much about MATLAB or the BCI competition, but I believe the approaches to increase accuracy are the same regardless of the platform used. First, what is the accuracy in your training data? and what deep learning method are you using?
1. Did you try to increase the sample size of your training data? I think this is the most effective way to get better results.
2. Did you try to increase the model size? perhaps you can get better results by adding more layers. If the accuracy in your training data is far higher than the accuracy in the testing/validation data, then you have to do the exact opposite (reduce the number of layers).
3. You can achieve higher accuracy by implementing some form of feature (predictors) engineering.
4. Finally, perhaps you can adjust some of the parameters used in deep learning like for example the learning rate. You can try to apply some form of regularization or simply add some dropout layers. This will work the best if you have overfitting (worse performance in the testing data when compared with that of the training data).
Hi!
First of all, I am not a Matlab expert (just know some basics, so I cannot help you much with the code itself). Secondly, you didn't mention what feature selection algorithm you used and which method (I guess some machine learning algorithm) did you use for the label prediction/classification.
After little search I saw you have: https://www.mathworks.com/help/stats/feature-selection.html all this methods available for feature selection; since I am not familiar with thet data , so for general purposes the random forest/decision trees /ensemble of bagged decision trees are often very good, but of course it depends on the data.
Besides, you also have Adaboost in Matlab: https://www.mathworks.com/discovery/adaboost.html and here you also have an explanation when to use each type of boosting algorithm: https://www.mathworks.com/help/stats/framework-for-ensemble-learning.html#bsw8akh .
I hope this helps you!
One way to improve the performance of a model that does the classification is to train your model with more data. Because basically, misclassification means the features of this particular sample haven't been learnt by this model. If you don't have more data to re-train your model, think about those helpful techniques, such as "data augmentation" and "Transfer Learning". They work pretty well with small datasets.
If You want to decrease the misclassification just balance your samples in each class. And if u want to increse the accuracy just take very small value for initial learning rate while defining options parameters.
First, you should compare the accuracy of training, validation and test data.
If the training accuracy is low, the model doesn't deal with the complexity of data, then you should adjust the parameters.
if the validation accuracy is low, it means that your model is overfitting
if the test accuracy is low, it means you need more diverse training data
- Badges
- Science topic