Time Series Data Reconstruction

#Help_post I have some time series data with some random missing values(values are deleted at random places and 10%,20%,30%.....,70%,80% data are deleted and thus some data of size 5000 is created.) I need to reconstruct the data back to the original one. In recent years CS(compressed sensing) algorithm made better reconstruction with PSNR value between 20-50dB(the higher the data percentage deleted, the lower the PSNR). I need to do the same stuff with ML or DL algorithm. I tried general LSTM or RNN algorithms, MICE algorithm, general sequential time series algorithm etcs. But it didn't give me better reconstruction. What can I do now?