Segregating Biased Data in Healthcare Datasets using MATLAB

We attempt to detect & display bias in datasets using MATLAB. Our prospects extend to using an embedded C++ file and having a webpage that allows users to test the quality of their data.

Comment

Mapping Skills to Projects

Our group had the following skills when it came to hacking, _ C++, C, SQL, Python, HTML & CSS, MATLAB, AI-ML, Image Processing, and Simulink_. We are a total of 4 of which 2 are biotechnology and 2 are computer science students. It was only ideal to pick AI in Healthcare.

Decoding the Problem

This was our problem statement AI models used in healthcare are built on previous data. But, when both medical knowledge and treatment are biased and used as data fed into AI models, we end up with AI that exacerbates current and past issues and prejudices within healthcare. How can we make sure AI models account for biased data?. We broke it down into 2 requirements, Healthcare-related dataset and Use AI to check for bias in the dataset.

Approach to Solve

We decided to use Excel and MATLAB to solve this problem. Excel will be used to create the dataset, and MATLAB will allow us to detect the bias in the dataset if any. Our main objective was to look for trends in the dataset. With an assumption of 80% of the data being accurate, we could use simple math to figure out any biases.

Algorithm

The logic was to find a trend and see if each data follows the trend. If any data was too vaguely out of trend, then it would be considered a biased dataset. This may not be the most effective solution, but with our time constraints, we had to come up with something fast.

''' if dataset(i) >> trend or dataset(i) << trend biased else unbiased '''

Implementation

The above algorithm looks so simple, but our inexperience in MATLAB took us an entire 4 to 5 hours to come up with a working code. We had ideas to embed a C++ code that when connected to the website built with HTML will allow users to check for the validity of their data. Due to this setback, however, we had to settle with just developing a C++ code.

Execution

In the end, we were able to put together 3 codes with a video demonstrating the entire working, PPT, a word document, built an entire Dataset, made a thumbnail (ofc), and uploaded them to the GitHub repo.

Built With

Share this project:

Updates