Contents
What is the advantage of the linear classifier?
The major advantage of linear classifiers is their simplicity and computational attractiveness. The chapter starts with the assumption that all feature vectors from the available classes can be classified correctly using a linear classifier.
Which type of problems Cannot be solved by a linear classifier?
It is clear that drawing one straight line cannot identify all the points of one of the class correctly. The above problems are called nonlinear classification problems and cannot be solved by drawing a linear classifier; therefore, other alternatives are required.
Is linear regression a classifier?
Linear regression is suitable for predicting output that is continuous value, such as predicting the price of a property. Whereas logistic regression is for classification problems, which predicts a probability range between 0 to 1. For example, predict whether a customer will make a purchase or not.
How are linear classifier different from non-linear classifiers?
The above problems are called nonlinear classification problems and cannot be solved by drawing a linear classifier; therefore, other alternatives are required. We may need piece-wise linear (i.e. linear in parts), or non-linear classification boundaries to identify the two classes correctly.
How can we build non-linear classifiers intuitively?
But the obvious weakness is that if the nonlinearity is more complex, then the QDA algorithm can’t handle it. For example, if we need a combination of 3 linear boundaries to classify the data, then QDA will fail. Logistic regression performs badly as well in front of non linearly separable data. We can see the results below.
Why do we need non-linear classification boundaries?
We may need piece-wise linear (i.e. linear in parts), or non-linear classification boundaries to identify the two classes correctly. What would be that non-linear function (or approximation) and how that boundary may look like is defined by a non-linear classifier.
How can we build non-linear classifiers with LDA?
In conclusion, it was quite an intuitive way to come up with a non-linear classifier with LDA: the necessity of considering that the standard deviations of different classes are different. But the obvious weakness is that if the nonlinearity is more complex, then the QDA algorithm can’t handle it.