A Taxing Problem

Project Info

Nostradata thumbnail

Project Description


Our focus was the challenge of identifying where the ATO should locate their Tax Help Centers. This project also addressed the challenge of combining data sets and using open data to help governments answer questions.


Data Story


We set out to demonstrate in this challenge how Machine Learning can be used to help government agencies make decisions, in this case where to put Tax Help Centers to best serve the community.

We utilised this opportunity to merge data from a variety of sources, namely:

1. ATO individual tax return data from 2014 - 2015 & 2015 - 2016
2. ABS demographic summary data from 2015 & 2016
3. Tax Help Center locations
4. ABS Geography publications

Given that the current distribution of Tax Help Centers is not necessarily optimal, the merged data was used to calculate an adjusted score for each postcode's requirement for tax assistance based upon the ATO eligibility criteria.

The adjusted scores were used to train a deep neural network to predict the required number of Tax Help Centers in a postcode, which after using cross-folds validation to verify the accuracy of the model, achieved a score of 96%


Evidence of Work

Video

Homepage

High-Res Image

Team DataSets

Taxation Statistics 2015-16

Data Set

ATO GovHack 2017

Data Set

ABS Geography Publications

Data Set

ATO GovHack 2018

Data Set

Taxation Statistics 2014-15

Data Set

Challenges

Bounty: Mix and Mashup

Region: Australia

Challenge

More than apps and maps: help government decide with data

Region: Australia

Challenge

Bounty: Tax Help Centers

Region: Australia

Challenge
Back to Projects