I am working on Continual learning with tabular data (using Keras, pandas, Tensorflow.
I have trained the model on the tabular data using MLP, where I have an Accuracy of 98% (I am using K-fold for cross validation in MLP Model) for time being it is called as pre-trained model and we are saving the model as ‘model.h5’ file
In second step, we are loading the pre-trained model to predict the new incoming tabular data in which we are getting an accuracy of 94%
As accuracy has been dropped from 98% to 94%, we are fitting the new data to model using model.fit()
When we try to print accuracy after fitting the new data, the accuracy has been dropped to 70%.
I. Does model.fit() actually means we are training the new data without forgetting the old data.
II. Why the accuracy has been dropped to 70% after training also on new data?
What can be done to avoid these issues?
Snippet of the code, the issue which is mentioned above:
