Feedforward Neural Network

This repository trains and evaluates a feedforward artificial neural network (FFNN) to predict whether a customer will subscribe to a term deposit using bank marketing campaign data. A logistic regression model is also included as a baseline.

What this project does

• Loads the dataset (as1-bank.csv)
• Applies basic pre-processing (encoding and scaling)
• Tests different feature selection options
• Trains multiple FFNN architectures (different layer sizes)
• Evaluates performance using metrics suitable for imbalanced classification
• Exports results to CSV and Excel

Dataset

The target label is y (subscription outcome). The classes are imbalanced (more “no” than “yes”), so metrics beyond accuracy are used.

Pre-processing

The script:

• Converts yes/no fields (e.g., default, housing, loan, and y) into 1/0
• Splits into train/test sets using stratification
• Scales features using StandardScaler

Feature selection options

The following feature sets are evaluated:

• All features
• Variance threshold (removes very low-variance features)
• Correlation filter (removes highly correlated features)
• Combined (variance + correlation)

Models

Baseline: Logistic Regression

• Uses class_weight='balanced' to account for class imbalance.

Feedforward Neural Networks (TensorFlow/Keras)

• Dense networks with ReLU hidden layers and a Sigmoid output layer
• Uses class weights to reduce bias towards the majority class
• Uses early stopping to help prevent overfitting

Architectures tested include:

• Single hidden layer: [32], [64]
• Two hidden layers: [64, 32], [128, 64]
• Three hidden layers: [128, 64, 32], [256, 128, 64]

Evaluation

Metrics reported include:

• Accuracy, Precision, Recall, F1-score
• MCC (Matthews Correlation Coefficient)
• Balanced Accuracy, G-Mean
• AUPRC (Area Under the Precision–Recall Curve)
• Confusion matrices for top models

Outputs

Running the script generates:

• model_comparison_results.csv
• model_comparison_results.xlsx

How to run

1) (Recommended) Create a virtual environment

python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

2) Install dependencies

pip install -U numpy pandas matplotlib seaborn openpyxl scikit-learn tensorflow keras

3) Run the script

python feed-forward-ANN.py

Repository contents

• feed-forward-ANN.py — pipeline: pre-processing, feature selection, training, evaluation, and exporting results
• as1-bank.csv — dataset used by the script
• model_comparison_results.csv — exported comparison table
• model_comparison_results.xlsx — exported comparison table (Excel)

Happy to provide my report on this repo if requested, reach out to oscar.a-r@hotmail.com