Improve accuracy of probabilistic linear regression

How can I improve the accuracy of my linear regression model ?
I have created a model to fit my data. when I extracted the results using sample_ppc and plotted it against my actual posterior, it looks fine. But, when I tried the same model with new data, it had a lot of deviation.
image
Initial Data
image
New Data

Green line is the actual data and red colored one is the predicted data.
It would be helpful if someone helps me out to increase the accuracy of my model

It looks like the you’re working with a model that just is not appropriate for the data. Your predictions on the new data match what the model learnt from the training data (an almost constant output centered around 0.25). but just don’t fit the new data well.