How to reduce Bias in Machine Learning
Hello fellow machine learning enthusiasts, today we are going to learn about how to reduce Bias in Machine Learning.
Well, we all have reached the stage, where even after trying every rule in the book, the accuracy just doesn’t seem to increase. So, let’s just try something new, what about reducing the bias.
Many times the problem is that the algorithm we implemented has a high bias.
But, what is this “bias” ????
What is a “bias”????
Bias is basically how well our model predicted the value over the actual value.
A high bias condition will lead to the algorithm missing the patterns or would not be able to accommodate the relation between the input and the output. When the bias is too high, it is assumed that the model implemented is way too simple and is not able to accommodate the complexity of the features and the labels and hence is not able to determine the relation between them correctly and hence leads to underfitting.
What does a “High Bias” imply????
- Underfitting: A model with a high bias is much simpler to accommodate the data set, hence it tends to underfit. That is the model is not able to learn and determine the required pattern.
- Low Training Accuracy: A model with a high bias does not fit the model very well hence it leads to a low training set accuracy.
- Inability to solve complex problems: A model with a high bias is unable to learn the complexity of the dataset because it’s a simple model.
In the figure above, the data points are soo far from the decision boundary, hence depicts a case of underfitting.
Yaaa…. you might be thinking “That is a really big issue”.
But, How do we solve this????
Reducing The Bias:
Following are the best practices to reduce bias:
- Adding neuron layers or input parameters for complex problems, adding neuron layers will help in increasing the complexity of the model while training, hence making it much better at fitting the complex dataset, which will increase the training accuracy.
- Adding more training samples, adding more training examples will help in representing more variance of parameters, hence the model will train better.
- Decreasing regularization parameter, if you would have implemented regularisation to improve the accuracy of the model, then you might have overdone it! Just try decreasing the ratio to improve accuracy and reduce bias.
- Train Longer.
And that’s how you reduce bias in machine learning.
I hope you learned something new today.
Thank you for reading.