Bias and Variance in Machine Learning
Bias and Variance are the parameters related to predictions of a neural network. Based on the accuracy of the prediction we can say whether the trained model is useful or not. But before that we need to know what exactly these terms are, So let’s jump into it.
Bias:
Bias is nothing but difference between the ground truth which we are expecting to be predicted by a model and the average prediction that our model gave on particular data.
A Model with High bias pays very less attention on training data and in result the performance on training and testing data is very poor. This leads to Underfitting problem. As this oversimplifies the problem, Model is unable to learn and generalize well.
Variance:
Variance, As the name suggests it tells us about how spread our model prediction data is. The Model with High Variance is able to fit the training data well but it is unable to make predictions on the unseen data. So, The Accuracy on training data is good but when it comes to testing on unseen data the accuracy will be low as this model with high variance can’t able to generalize well.This leads to Overfitting problem.
Don’t worry if you’re unable to understand these parameters,A figure will be added to justify these things which makes more meaningful to understand.
You Can have a look into my previous article if you wanted to know about Overfitting and Underfitting. Here’s the Link
https://saichandra1199.medium.com/underfitting-and-overfitting-in-machine-learning-e8c21c2611c7
This is known as Bulls-Eye Diagram to represent all possible cases of Bias and Variance at once for better understanding.
Here the Central part represents the Ground Truth which we want our model to be predicted. That can be attained only with Low Bias and Low Variance. Hope you got some essence of bias ,variance and Underfitting and Overfitting as well.
Bias and Variance Tradeoff:
If our model has large number of parameters it is complex and has high variance and low bias.On the other hand,If our model is too simple and has very few parameters then it may have high bias and low variance. An algorithm can’t be more and less complex at the same time so it is not good to have both bias and variance high. So we need to find the right/good balance without overfitting and underfitting the data.
So make the model perform better understanding these terms is much critical and useful. Total Error considering bias and variance is given below
Total Error = Bias² + Variance + Irreducible Error.
Irreducible error is the error that can’t be reduced by any model. As it is a measure of the amount of noise in our data. Because no matter how good our data is, It will contain certain amount of noise or irreducible error that can not be removed.That’s it for this title.
Thanks for reading my article!!!! Do Follow for more interesting stories :)
