FairLens-AI: A Comprehensive Solution for Machine Learning Fairness Detection and Bias Mitigation

FairLens-AI is an open-source machine learning fairness auditing platform that provides end-to-end tools from data upload, bias detection to mitigation strategies. It integrates Google Gemini AI to generate intelligent analysis reports, helping data scientists and engineers identify, quantify, and mitigate biases in models, and lowering the technical barrier to fairness auditing.

In the era of widespread AI application, machine learning models influence key decisions such as loan approval and recruitment screening. If training data contains biases, models will perpetuate or even amplify unfairness, harming the interests of specific groups and potentially exposing enterprises to legal risks and reputational damage. FairLens-AI is an open-source project created to address this issue, providing an end-to-end fairness auditing platform.

FairLens-AI is a web-based interactive platform with a front-end and back-end separation architecture: the back-end uses FastAPI to build high-performance API services, and the front-end uses React + Tailwind CSS to provide a modern interface; it integrates Google Gemini 2.0 Flash to generate intelligent reports; the deployment architecture includes Google Cloud Run (containerized back-end) and Netlify (static front-end hosting). Core functions cover data upload and analysis, fairness metric evaluation, bias detection, mitigation strategies, and AI report generation.

FairLens-AI implements industry-standard fairness metrics: demographic parity difference (ideal value <0.10), disparate impact ratio (compliance threshold ≥0.80), equal opportunity difference; it automatically detects bias types such as class imbalance, representational bias, and outcome disparity, and classifies risks into low/medium/high levels.

For detected biases, it provides three mitigation methods: reweighting (adjusting sample weights), resampling (oversampling minority groups), feature removal (deleting sensitive attributes), all with visual comparisons before and after mitigation; it integrates Google Gemini to generate context-aware audit explanations, bias reasoning analysis, technical improvement suggestions, executive summaries, and downloadable PDF reports.

FairLens-AI is applicable to scenarios such as financial credit (loan approval compliance), recruitment screening (resume screening fairness), medical diagnosis (disease prediction consistency), and judicial risk assessment (racial bias-free recidivism prediction). To get started quickly, you can upload a sample dataset (loan approval scenario), select sensitive attributes (e.g., gender, caste) and target columns (e.g., loan_approved), run the analysis, view results, apply mitigation strategies, and download reports.

FairLens-AI provides a fully functional, easy-to-use open-source solution that lowers the technical barrier to fairness auditing, enabling non-technical personnel to understand model biases. As global AI regulatory regulations advance (e.g., the EU AI Act), such tools will become standard components in machine learning engineering. The project is open-source under the MIT license with a clear and extensible code structure; it is recommended that teams building or deploying machine learning systems include it in their toolkits.