I modeled a multi class predicting neural network I provided train data (800 entries)and prepared model for prediction. Now I used test data (100 entries) and predicted there corresponding label correctly. Is it possible to get a prediction for a single user input. Hand written digits prediction dataset, I loded from keras, I am unable predict the hand written digits captured from mobile camara and reduced to required pixel ratio and converted to gray scale then scaled and flattened then supplied to neural network.
You can look into this
https://machinelearningmastery.com/loocv-for-evaluating-machine-learning-algorithms/
800 entries are pretty low for such task.
800 - 100 means your network has been overfitted to that particular dataset.
Why did not you tried MNIST dataset and add more training data?
Then you may need to fine tune the model for your domain (for data from your setting).
But first try to add more data to training set.
You need to apply the pipeline that you created using your training set to the input sample you want to predict. For instance, if you only considered standardization you would do the following:
1. Load the user input (an array with dimension 1 x number of features).
2. For each column in the array, remove the corresponding training mean and divide by the corresponding training standard deviation.
3. Estimate the prediction for the standardized input array.
- Badges
- Science topic