At least the number isn't entirely arbitrary picked.
2. How many previous block should be considered to calculate average transaction fee?
One.
3. How often should block limit changed?
On every block.
The number you chose have some obvious problem, have you considered the possibility of
1. Miner mine 2 blocks at once, where the 2nd block usually are empty.
2. Analysis which shows certain days allows you move Bitcoin quickly with lower fees. See
https://bitcointalk.org/index.php?topic=5250569.0.
3. Messing with various transaction fee estimation algorithm, which leads user pays too much or waiting too long.
Ok, we must not forget that there is some electricity cost for mining blocks, mining an empty block is not profitable. An adjustable blocksize cap algorithm would reduce the limit of the block when the demand for space is lower which would maintain some level of competition for block space (we could modify the min limit if needed). I believe that the transaction fee market will barely react to the size of the next block as it would propably change all the time but we know for sure that the processing speed would increase in periods where the demand for space is higher.
I think that an immediate adjustment to the change in price would be the best...
You can't include external data without oracle or someone who you must trust.
I wasn't talking about importing data from external sources. You can easily compute the avg transaction fee on the last block:
total amount of all the transactions inputs - total amount of all the transactions outputs quantity of transactions
@amishmanish, I think this post may have addressed your concerns as well...