This repository contains the final project submitted for the AIM 846: Multiobjective Machine Learning course.

• Niranjan Gopal
• Akash Sridhar

The project investigates the trade-off between fairness and accuracy in machine learning models through multi-objective optimization techniques. We aim to identify Pareto-optimal solutions that balance predictive performance with fairness constraints.

We utilize the Bank Marketing Dataset, which contains information about direct marketing campaigns of a Portuguese banking institution. The sensitive attribute considered for fairness evaluation is marital.

We used the following solvers for multi-objective optimization:

• Optuna
• BoTorch

• 1_eda_preprocess.ipynb: Data exploration and preprocessing steps
• 2_optuna.ipynb: Hyperparameter tuning using Optuna
• 3_botorch_lr.ipynb: Bayesian optimization for Logistic Regression
• 3_botorch_nn.ipynb: Bayesian optimization for Neural Networks
• 3_botorch_nn2.ipynb: Cleaned-up NN notebook with GPU/CPU workflow handling
• 3_botorch_rf.ipynb: Bayesian optimization for Random Forest
• 4_combining_pareto_from_all_models.ipynb: Aggregating Pareto fronts from all models
• bank.csv: Original dataset
• processed_bank_marketing.csv: Preprocessed dataset
• each_models_pareto_results.csv: Pareto results for individual models
• final_pareto_front.csv: Combined Pareto front
• requirements.txt: List of dependencies

1. Clone the repository:

   git clone https://github.com/Niranjan-GopaL/Modeling_Fairness_Accuracy_MOO_approach.git

2. Navigate to the project directory:

   cd Modeling_Fairness_Accuracy_MOO_approach

3. Install the required packages:

   pip install -r requirements.txt

Run the Jupyter notebooks in the order listed above to reproduce the results. Ensure that the datasets are placed in the appropriate directories as referenced in the notebooks.