Riol26

Challenge 3: Modelling Energy Consumption & Fairness of Errors

AGUILAR, COUALLIER, DUPUY, ST EXUPERY

Description

We built predictive models to forecast daily energy demand and evaluated their fairness across building types.

• Modelling: Engineered lag features and compared Linear Regression vs. Random Forest. Both models showed high accuracy ($R^2 \approx 1.0$) due to the nature of the interpolated data.
• Fairness of Errors: We calculated the Mean Absolute Error (MAE) for different groups. We analyzed whether the models performed equally well for Public Schools compared to Residential Homes to avoid discriminatory precision in energy management.

How to run

Click the badge below to open the notebook directly in Google Colab.

Riol26

Challenge 2b: Fairness Metrics & SQL Sampling (AIF360)

AGUILAR, COUALLIER, DUPUY, ST EXUPERY

Description

We focused on detecting and mitigating bias in a socio-economic dataset before using it for AI training.

• Tools: IBM AI Fairness 360 (AIF360) and SQLite.
• Metrics: Calculated Disparate Impact and Statistical Parity Difference on the Adult Census Income dataset.
• Mitigation:
  1. Applied the Reweighing algorithm to correct bias in the training data.

2. Implemented Balanced Sampling using SQL window functions (ROW_NUMBER()) to ensure equal representation of groups.

Riol26

Riol26