Inspiration

It has always been a struggle for companies to improve the efficiency of their marketing strategy. We believe that there is a simple way for a company to improve its ad services without sharing its valuable user data with giant online tech firms like Google and Facebook, which are under more and more scrutiny by the public and legislative regulatory for privacy concerns.

What it does

Our Decision Tree Model is capable of accurately telling the difference between a customer of the high potential of making an in-app purchase of a mobile game with the other users.

Our associated website provides an interactive interface to upload datasets and demo the appropriate predictions. It also provides additional information such as a suggestion for the system to provide a moderate discount to encourage customer with purchasing potential to even spend more on the game.

How we built it

We used ID3 (Iterative Dichotomiser 3) Decision Tree Learning to train and test the data extracted from the raw information provided.

We used the mean feature values to find the value to fill the empty NaN slots for some customer information to train the decision tree. In the end, we fine-tuned the model by selecting relevant features to prevent garbage data influences the model.

Challenges we ran into

Creating the web-app and the ML model required very different skill sets and technologies which we weren't at first familiar with.

For the decision tree model, we initially didn’t know how to implement it in our project as a BlackBox. We obtained some inspirations form the famous Titanic survival model link.

Accomplishments that we are proud of

We finished the project on time with learning a lot of the inner working of the decision tree! The Model runs and is accurate and the website successfully built.

What we learned

A lot of decision tree learning and python knowledge.

Also, some web development frontend/backend implementation.

What's next for GameTreeAI

There are many ways the model can be improved:

  1. Only a binary decision?
  2. ID3? CART, C4.5?
  3. Filtering garbage data?
Share this project:

Updates