I want to make a model to predict the out come of a game while it is still being played. My data consist of the result of the past games and some simple information about the state of those games such as scores and positions of each player. The information are taken every minute until the end of the game. I have many sample for the game where each sample is the result + info about the game perminute. the max minute is 60 so if the game end before 60 minutes then the data will just be the initial value 0.

Time serie model I searched up online all points to single time line, they forecast based on the past, but my want my model to forecast based on the progression of the game including what happen in the past.

I will use my model while a game is going on and it will update the chance of winning every minutes. So the input is only related to previous data up until some point in time.

I tried using lstm but it does not seems to work well. I was expecting it to be learning my data like a language model where it tries to predict the game state from past to current information but the accuracy does not go beyond 55%. I tried tuning the parameters but does not seems to make a huge difference.

What other models can i use for this? Or what clever methods can I use?