https://github.com/barrettotte/anilist-ml
Training a binary classifier model to predict if I would recommend an anime using my Anilist user data.
https://github.com/barrettotte/anilist-ml
anilist binary-classification data-visualization machine-learning scikit-learn
Last synced: about 1 month ago
JSON representation
Training a binary classifier model to predict if I would recommend an anime using my Anilist user data.
- Host: GitHub
- URL: https://github.com/barrettotte/anilist-ml
- Owner: barrettotte
- License: mit
- Created: 2022-09-17T23:40:07.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2022-10-05T18:25:26.000Z (over 3 years ago)
- Last Synced: 2025-07-16T22:18:59.496Z (11 months ago)
- Topics: anilist, binary-classification, data-visualization, machine-learning, scikit-learn
- Language: Jupyter Notebook
- Homepage:
- Size: 25.5 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# anilist-ml
Learning the basics of machine learning with Anilist data
## Summary
I originally set out to make a model trained on my Anilist data that was able
to predict the score (1-10) that I would probably give an anime.
At first I tried to make a regression model. I realized since I was only rating 1-10
I would probably have more accuracy using a classifier model. Classifier models were definitely more accurate.
But, the best I was able to pull off was a random forest classifier with an accuracy of ~38%.
I think my data is subpar since I only have ~500 data points and generally skewed data.
I tend to give an anime a score of 7 more than anything else and also tend to not be very critical.
So, entries with a score below 5 are very rare and I think it makes training more difficult.
After messing around for a while I decided to switch to a binary classification and change my target.
Would I recommend or not recommend an anime?
With this change, my final result was a random forest classifier with an accuracy of ~84-85%.
Someone knowledgeable in machine learning could probably point out what I was doing wrong immediately.
But, oh well this was my first ML project and I have a lot to learn.
This wasn't exactly a win, but hopefully the next ML project goes better.
## Results
The final model is located in [model.ipynb](model.ipynb).
```txt
Log loss = 4.934196584711905
ROC AUC Score = 0.8524492234169653
Classification Report:
precision recall f1-score support
False 0.826087 0.612903 0.703704 31
True 0.865169 0.950617 0.905882 81
accuracy 0.857143 112
macro avg 0.845628 0.781760 0.804793 112
weighted avg 0.854351 0.857143 0.849922 112
```
ROC Plot
### Notebooks
1. [fetch.ipynb](fetch.ipynb) - Fetches user and anime data using Anilist's GraphQL API.
2. [clean.ipynb](clean.ipynb) - Cleans fetched data to prepare for data visualization and model training.
3. [explore.ipynb](explore.ipynb) - Some data visualizations and general exploration of data.
4. [model-select-reg.ipynb](model-select-reg.ipynb) - Testing out different regression models.
5. [model-select-cls.ipynb](model-select-cls.ipynb) - Testing out different classifier models.
6. [model.ipynb](model-final.ipynb) - Final model training, verification, and evaluation.
## Data
- fetched
- [data/anime-YYYYMMDD-raw.csv](data/anime-20220927-raw.csv) - raw Anilist anime data
- [data/user-YYYYMMDD-raw.csv](data/user-20220927-raw.csv) - raw Anilist user data
- cleaned/enriched
- [data/anime-YYYYMMDD-clean.csv](data/anime-20220927-clean.csv) - cleaned anime data; usesless columns dropped and missing data filled
- [data/user-YYYYMMDD-clean.csv](data/user-20220927-clean.csv) - cleaned user data; useless columns dropped
- [data/user-YYYYMMDD-enriched.csv](data/user-20220927-enriched.csv) - user data joined with anime data
- regression
- [data/user-YYYYMMDD-reg-train.csv](data/user-20220927-reg-train.csv) - train data for regression models
- [data/user-YYYYMMDD-reg-valid.csv](data/user-20220927-reg-valid.csv) - validation data for regression models
- [data/user-YYYYMMDD-reg-test.csv](data/user-20220927-reg-test.csv) - test data for regression models
- classification
- [data/user-YYYYMMDD-cls-train.csv](data/user-20220927-cls-train.csv) - train data for classifier models
- [data/user-YYYYMMDD-cls-valid.csv](data/user-20220927-cls-valid.csv) - validation data for classifier models
- [data/user-YYYYMMDD-cls-test.csv](data/user-20220927-cls-test.csv) - test data for classifier models
I also made the Anilist anime data (as of 09/27/2022) available on
[Kaggle](https://www.kaggle.com/datasets/barrettotte/anilistanimedata).
## References
- [Anilist Interactive GraphQL Tool](https://anilist.co/graphiql)
- [Anilist GraphQL Documentation Explorer](https://anilist.github.io/ApiV2-GraphQL-Docs/)
- [Coursera Machine Learning Specialization (Andrew Ng)](https://www.coursera.org/specializations/machine-learning-introduction)
- [Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow. Geron](https://www.amazon.com/Hands-Machine-Learning-Scikit-Learn-TensorFlow/dp/1492032646)