My predicted ratings

Created by kvib.

Favorited 1 time, disliked 0 times, added to 0 watchlists.

I applied machine learning to my movie ratings at Letterboxd to predict rating scores for movies on some of the main official lists on ICM.

For input features I used ratings, likes, view counts and some minor metadata like release year, run time and genres from the following sites:
* ICheckMovies
* IMDb
* Letterboxd
* Rotten Tomatoes
* Metacritic

The used algorithm was linear XGBoost, I imputed any missing ratings (especially for older movies on Metacritic).

The most important features turned out to be:
icm_favs 80.30
letterboxd_rating 35.15
letterboxd_likes 34.59
imdb_rating 7.35
rt_critic_score 7.17
icm_checks 6.59
metacritic_user_score 5.69
metascore 5.39
rt_critic_count 5.35
runtime 5.22
year 5.13
rt_audience_score 4.09
icm_lists 3.74
imdb_votes 3.13
letterboxd_watches 3.04
letterboxd_lists 2.84
genre_romance 0.89
genre_crime 0.78
genre_fantasy 0.49
genre_thriller 0.36
genre_war 0.31
genre_adventure 0.30
genre_scifi 0.28
genre_mystery 0.17
genre_biography 0.11
genre_history 0.10
genre_animation 0.10
genre_documentary 0.10
genre_horror 0.06
genre_music 0.01
genre_western 0.01
genre_short 0.00

The predicted rating value was more than 4.5 for the top 70, 4 for the top 398, 3.5 for the top 587 and 3 for the top 670.