Basically, Ive come to the realization that trying to predict playoffs based on ranking individual teams in a single list is deeply flawed, because it does not take series matchups into account. What really needs to happen here is some sort of playoff series object which contains each team playing in the series, along with a few short pieces of info (seasonId, seasonIndex, round number, round number below finals, etc.)
The compared values of the two teams can then be used to determine what predicts the final series goal diff & games winner, penalty minutes, OT games, whatever, which can then be used to create a model (likely on a normal distribution) that shows the probability of each team winning a given playoff series.
Basically, Ive come to the realization that trying to predict playoffs based on ranking individual teams in a single list is deeply flawed, because it does not take series matchups into account. What really needs to happen here is some sort of playoff series object which contains each team playing in the series, along with a few short pieces of info (seasonId, seasonIndex, round number, round number below finals, etc.)
The compared values of the two teams can then be used to determine what predicts the final series goal diff & games winner, penalty minutes, OT games, whatever, which can then be used to create a model (likely on a normal distribution) that shows the probability of each team winning a given playoff series.