The Bias versus Variance Tradeoff
This was very confusing term, every time I read and forgot. This time me and even you will never forget because I learned some easy technique to remember. You can follow me along the easy step by step the trade off process.
- When call Bias ? High & Low
Bias is a measurement of how accurately a model can capture a pattern in a training dataset.
Train error variability called Bias.
- When call Variance ? High & Low
Test error variability indicate the variance
- When is come to Tradeoff?
When Training error & Testing error variability are low.
Bias and variance using bulls-eye diagram
For non linear model both sample 1 and 2 training error )0–0=0) is zero that means low bias and test error is (100–27=73) high difference that means high variance.
For Linear model you can observe that training error (40–27 = 13) and test error (45–34 = 11) difference is less compare to non linear, so it does mean that low bias and low variance.
To get balance model we should keep low bias and low variance.
To do that we can apply cross validation sample, regularization, dimensionality reduction, ensemble methods, bagging & boosting.
I hope my simple writeup will help you to remove confusing about bias variance tradeoff and this better understanding will help you to better perform in data science domain. Github (https://github.com/mamun21616), Linkedin..https://www.linkedin.com/in/abdulla-mamun-4222b11a/
