gabygutzp / targeted_debiasing

MSc Research - Feature targeting debiasing for Machine Learning applications in finance

Targeted_Debiasing

MSc Research - Feature targeting debiasing for Machine Learning applications in finance

In this thesis, we tackle the important research problem of algorithmic bias in Machine Learning (ML). A key issue that potentially compromises the fairness and transparency of predictive models, this bias is often embedded within specific features of the model. We focus on using feature importance as a tool to surface and reduce this bias, striving for a method that decreases the bias induced by individual features without sacrificing their predictive power. This focused approach represents a departure from traditional strategies that aim to debias data as a whole. We aim to mitigate the size of the typical accuracy-bias tradeoff that is common in algorithmic fairness research. By preserving the predictive power of the model while simultaneously reducing bias, we aim to achieve a balance of fairness and accuracy, a highly desired yet elusive goal in Machine Learning.

The code can be found in the notebook: Thesis_v2.ipynb

The full dataset: AdultData.csv Training data: clean_encoded_adult_train_v2.csv Test data: clean_encoded_adult_test_v2.csv