Estimating Access to Mobile Broadband using CNN
Equitable access to broadband is necessary for inclusion in the information economy.
As the world economy continues to transition towards the information economy, it is essential to have access to broadband. However, and estimated 22 million Americas do not have access to broadband. They are essentially cut-out of the future economy. To change this we need to understand where and how many people do not have access to broadband.
The current representation of this is through the FCC broadband data, which is notoriously inaccurate and severely under-represent the number of Americans without access. The collection of this data, and even more so more accurate data, is very difficult and costly. We wanted to use Deep Learning to use this collected data, satellite imagery and cell tower locations to estimate access. This would be able to provide similar mapping but at a much lower cost. For this project we implemented a CNN for image segmentation to train our model. We implemented this for 5 levels of access (0-20%,20-40%, etc.) over 6 states and got 97% and 80% categorical accuracy of the training and test sets, respectively. A link to our GitHub repo is here. Processes: data processing, deep learning, CNN, data visualization, python, Jupyter notebooks |