Decision Tree in Python : Metaprotein

Python code

R code

---

Decision Tree

We have implemented the following criteria to split nodes in our Decision Trees for comparison of accuracy

• Gini Index
• Information Gain (Entropy)

Metaprotein Dataset

Metaproteins as rows and, patient data of three types with samples from each being tested for the presence of metaproteins in columns (along with metaprotein demographics)

To suit our decision tree model, we removed the demographic columns from the dataset and have transposed the data frame to turn metaproteins into columns/variables & patients as rows.

We created a class label "Patient type" which has 3 factors - C, UC & CD

Metaprotein Dataset after pre-processing

We have taken half (1/2) of our Metaprotein Dataset to be used as Training Dataset & (1/2) to be used as Testing Dataset

---

Gini Index

• depth = 2
• leaf nodes = 4

Confusion Matrix: Prediction on Test Dataset

Accuracy = 19/24 = 79.16 %

---

Information Gain (Entropy)

• depth = 2
• leaf nodes = 4

Confusion Matrix: Prediction on Test Dataset

Accuracy = 17/24 = 70.83 %

---

Pruning

A tree that is too large risks overfitting the training data and poorly generalizing to new samples. We realized the necessity of pruning the decision trees by tuning parameters namely, "max_depth" and "max_leaf_nodes".

---

Pruned Decision Tree: Gini Index

• depth = 1
• leaf nodes = 2

Confusion Matrix: Prediction on Test Dataset

Accuracy = 54.1 %

---

Pruned Decision Tree: Information Gain (Entropy)

• depth = 2
• leaf nodes = 3

Confusion Matrix: Prediction on Test Dataset

Accuracy = 70.8 %

---