Python Artificial Intelligence Projects for Beginners

Support Vector Machine
 (
SVM
), which is the neural
network approach, the output is 29%:
The random forest is still better.
Let's perform cross-validation to make sure that we split the training test in different ways.
The output is still 44% for the random forest, 25% for our decision tree, and 27% for SVM,
as shown in the following screenshot:

---

Prediction with Random Forests
Chapter 2
[ 43 ]
The best results are reflected through random forests since we had some options and
questions with random forests.
For example, how many different questions can each tree ask? How many attributes does it
look at, and how many trees are there? Well, there are a lot of parameters to look through,
so let's just make a loop and try them all:

---

Prediction with Random Forests
Chapter 2
[ 44 ]
These are all the accuracies, but it would be better to visualize this in a graph, as shown
here:
We can see that increasing the number of trees produces a better outcome. Also, increasing
the number of features produces better outcomes if you are able to see more features, but
ultimately, if you're at about 20 to 30 features and you have about 75 to 100 trees, that's
about as good as you're going to get an accuracy of 45%.

---

Prediction with Random Forests
Chapter 2