Time Series Smoothing - Rolling Mean and Exponential Weighted Mean together?

Hi. I am not a stats person......!

I am analysing some Covid data for a Deep Learning project using Scottish COVID data. I have a question about smoothing the data.......

The deaths data is not reported at the weekend so a 7 day rolling mean seems essential to start off with:

Original Data:


7 Day Rolling Mean:

7 day Avg.png

However this still has some jagged bits! Not sure why. This means the learning algorithm doesn't work too well. So I have tried adding a second stage of an EWM firstly for 7 days:

7 day avg and 7 day EWM.png
And then for 14 days:

7 day average plus ewm.png

Now these seem to me to look much better at showing the trend and work better for the learning but I am worrying (perhaps unnecessarily) about my lack of rigour in understanding statistically what is happening.

Is there a better way of doing this?!

Thanks in advance.