You have got a dataset of 20 cars. Amongst them, 10 are luxury cars, and the opposite 10 are non-luxury. This dataset is represented within the shown below. A node is basically a containing members — on this case, cars. The green plus signs are luxury cars, and the red minus signs are non-luxury.
On this dataset, there are an equal variety of luxury and non-luxury cars. Such datasets are called .
The dataset has values of three features of every automobile.
- price
- fuel economy
- automation level
Example values of those features:
- Price — 12 lakh Indian Rupees. That is represented as 12L. This can be a .
- Fuel economy — there are 2 values for this feature — high and low. For instance, high will be fuel economy ≥ 15 kmph. Then low is fuel economy < 15 kmph. fuel economy is an (ordinal because there’s an order amongst its two values).
- Automation level — there are 6 levels of driving automation in a automobile, represented as L0, L1, to L5. L0 is the bottom, and L5 is the best. Allow us to assume the cars we’re considering on this problem will either be L1 or L2. So, this feature can be an .
Within the dataset, we’re given which cars are luxury and that are non-luxury. Our objective is:
- We will probably be given a automobile in the longer term with values for the three features. But, we is not going to know whether it’s luxury or non-luxury.
- We want to predict whether it’s a luxury or a non-luxury one.
This problem is known as . Because the dataset given to us is balanced, the classification is known as
As a primary step, allow us to divide the dataset into 2 subsets. The hope on this splitting process is to generate subsets in such a way that every subset is as as possible. What this implies is that the members inside each subset are as similar as possible to one another. The splitting process is finished using the three features.
Allow us to split the dataset using the value feature. Allow us to put the members which have a price ≥ 15L into one subset and people below 15L into one other.
The parent node has 20 cars, the left has 8, and the precise one has 12. Within the left child node, there are 2 luxury cars and 6 non-luxury. So, there’s a better percentage of non-luxury, i.e., many of the cars are non-luxury. In the precise child node, many of the cars are luxury. This is sensible because luxury cars will probably be dearer.
If a latest automobile with values for the three features is given to us, we see its price and ask an issue, which of the kid nodes will the brand new automobile fall into? Whichever class is almost all in that child node, we are going to predict that as a the category of the automobile. For instance, if the value of the brand new automobile is 18L, we predict that as a luxury one.
One in all the advantages we get from this splitting process is that we will be more confident in predicting a latest automobile as luxury/non-luxury using the kid nodes than using the alone. Within the parent node, there is no such thing as a clear majority, and we is not going to have the ability to say confidently that the brand new class is luxury or non-luxury. But for the reason that distribution of cars in each of the kid nodes is skewed towards one class, we are going to have the ability to say with more confidence that the brand new automobile belongs to the bulk class in whichever child node it falls under.
But, we have now performed the splitting process with one feature variable only. Now 2 questions arise:
- Why was the worth 15L chosen for the value to separate the dataset?
- What form of split will we get if we use either of the opposite 2 features?
Allow us to perform the split based on the fuel economy. The split looks like
Now, allow us to perform the split based on the automation level. The split looks like
The split based on automation level is sensible because luxury cars usually tend to have L2 automation than non-luxury ones. Similar reasoning applies to L1 automation.
We’ve created 3 different decision trees each based on one in all the three feature variables. In each tree, we have now split the dataset into 2 subsets based on a call. In decision trees, each node is a set of members within the dataset, and every edge is a call.
The two decision trees are shown below.
Which of those 3 decision trees gives us more confidence to us in predicting whether a latest automobile given to us in the longer term with values for the three features is luxury or not? Ideally, we’d have liked all luxury cars to be in a single child node and all non-luxury ones in the opposite node.
The choice tree created using the automation level feature is in a position to segregate luxury and non-luxury cars higher. The left child node during which the posh cars are present as the bulk has 80% of all members as luxury. But, within the 2 other decision trees, the proportion of luxury cars within the child nodes with the vast majority of luxury cars is lower than 80%. So, the choice tree based on automation level is best when it comes to because it has produced almost pure child nodes.
So, the expectation from a call tree in a classification setting is that it produces child nodes that contain one in all the classes with as high homogeneity as possible in order that we are able to make a prediction with high confidence based on which child node a latest datapoint falls into.
This decision tree-building process will be performed for and as well. I’ll explain the best way to mathematically perform the splitting process in future posts.
Signing off now!
