Exploring the rules of the decision tree using Graph objects

• Big idea -- Machine learning models are difficult to explain and interpret.
• Small idea -- Especially, in understanding the insights offered by the large decision tree models, such as, "How the rules affect the the entity" is challenging with complex models.
• Birds eye view of the idea -- We plan to undo the complexity of the large decision tree model and explain the relationship of between the insight within the model with respect to the data.
• Technical details -- Train a dataset on a simple machine learning model such as decision tree. Using graphs / networks understand the rules of the decision tree.
• What's next -- Often, the visualized network results in formation of various clusters. The future direction can explore the possibilities in identifying similarity between the structures.

🎯 Objective

• The objective of this research study is to develop a method for explaining large decision rules using a graph network and how it affects data samples.

• We specifically try apply this technique on the IRIS dataset.

• The objective of this research study is to develop a method for intepreting the large decision rules and how it affects data samples. We demonstrate our techniques with the use of IRIS dataset.

💪 Challenge

• In the real world dataset, it is a challenge to identify patterns and various factors that interact in explaining the decisions made for specific instance.

🧪 Solution

• With large set of complicated rules, it is difficult to understand the interactions.
• We use explainable methods such as decision trees.
• Identify the commonality between the entities by visualizing the rules between the trees.

🎬 Getting started

The following are the steps to setup this project:

Clone

git clone https://github.com/kaleoyster/decision-tree-viz.git

Run requirements.txt

pip install -r requirements.txt

Run the decision tree model

python3 src/decision_tree.py

Run http server

python -m http.server

View visualization

Serving HTTP on :: port 8000 (http://[localhost]:8000/) ...

👉 Additional references

Document	Documentation type	Description
Quickstart	Concept	An overview and guide to setup this project
Methodology	Concept, Task	Simplest possible method of implementing your API
Functions	Reference	List of references for the functions used
Related Projects	Reference	List of projects related to this repository