I analyzed data of 130 animes from studio SHAFT scraped from Myanimelist.net and built a linear model to try and predict user ratings/scores of future SHAFT animes.

R packages:

• core tidyverse
• gridExtra
• rvest

Data Source: https://myanimelist.net/anime/producer/44/Shaft

processed_data.csv columns:

• Title (Anime title)
• Eps (Number of Episodes)
• Src (Original Source of Anime)
• Genres (Genres, Separated by Comma)
• Members (Number of People with The Anime in Their MAL List)
• Score (User Rating from MAL Users)

Here are some results of the exploratory data analysis. The full analysis can be seen [here].

Because of the moderate positive correlation between the number of members and the user score, I tried a linear regression model. I also split the data into training and testing set with a 80:20 ratio. I tried 2 models, 1 with a log-transformed Members variable and 1 without. Here are the performances of the models:

Without transformation: MAPE = 8.51 %

With transformation: MAPE = 7.50 %