機械学習でデータ数がとても少ない時の処方箋

機械学習でデータ数がとても少ない時の処方箋として、linear modelがおすすめとの事。

Overfitが心配な場合、lassoなどが向いているかもしれません。

ただし、説明変数の厳選もまた大事。多すぎては良くない。

https://blog.bigml.com/2019/03/20/bigger-results-from-smaller-data-with-linear-regression/

https://www.kaggle.com/discussions/questions-and-answers/530115

https://qiita.com/suzuki0430/items/e3958f027af1dd011f21

https://qiita.com/c60evaporator/items/d0356fca12b37a82fe57

https://stackoverflow.com/questions/74922485/feature-importance-with-elastic-net

https://medium.com/@abhishekjainindore24/elastic-net-regression-combined-features-of-l1-and-l2-regularization-6181a660c3a5

https://mori-memo.hateblo.jp/entry/2022/02/22/011426#google_vignette

https://alok05.medium.com/ridge-lasso-and-elastic-net-regression-48a6684b7ead

https://ja.m.wikipedia.org/wiki/多重共線性

https://zenn.dev/tatamiya/articles/eeb97aae490f3ee8f0e9

https://qiita.com/Ogin0pan/items/4105c0c39869d31f4a75

https://zenn.dev/wsuzume/articles/fab5000ed8bafd

https://www.salesanalytics.co.jp/column/no00163/#Elastic_Net

https://datascience.stackexchange.com/questions/55831/why-does-lasso-behave-erratically-when-the-number-of-features-is-greater-than