Bank Telemarketing Success Prediction Using ANN

🎯 Project Description

A production-ready deep learning pipeline predicting bank telemarketing success via an ANN model trained on the UCI dataset with 82% accuracy[cite: 1, 2]. Features a dynamic Streamlit UI and FastAPI backend optimized with Pydantic validation to enforce strict input data boundaries and prevent inefficient model resource token burn. By processing complex client profiles alongside crucial macroeconomic indicators, the architecture replaces raw scripts with a resource-efficient inference application.

🖼️ Application User Interface

Live Preview of the Dashboard
(Once you upload your screenshot to GitHub, replace 'screenshot.png' with your actual image file name)

📊 Model Accuracy & Performance

• Model Framework: TensorFlow / Keras (Artificial Neural Network - ANN)[cite: 2]
• Target Objective: Binary Classification (Predicting yes / no for term deposit subscription)[cite: 1]
• Validation Accuracy: 82% 🔥
• Key Insight: The integration of 5 social and economic contextual attributes (such as employment variation rate and consumer price index) significantly enhances predictive performance compared to using traditional demographic features alone[cite: 1].

🧠 Neural Network Architecture

The deep learning architecture is constructed with the following layers:

1. Input Layer: Designed to receive preprocessed feature vectors derived from the 20 core dataset attributes[cite: 1].
2. Hidden Layers: Multiple Dense layers utilizing ReLU activation, combined with Dropout regularization to mitigate overfitting during training.
3. Output Layer: A single continuous Dense neuron configured with a Sigmoid activation function to compute precise probability scores between 0 and 1.

🛠️ Data Preprocessing Pipeline

To prevent data leakage and ensure 100% feature consistency during live inference, the pipeline utilizes a robust preprocessing setup:

• Categorical Encoding: Multi-class categorical variables (e.g., job type, education, marital status) are transformed using OneHotEncoder[cite: 1].
• Numerical Normalization: Continuous numeric attributes (e.g., age, consumer confidence index, euribor 3 month rate) are scaled using StandardScaler[cite: 1].
• Pipeline Synchronization: Transformations are bundled into an SKLearn ColumnTransformer, loaded via preprocessing.pkl and strictly aligned with live inputs using column.pkl[cite: 2].

💻 Installation & Setup

Run the following commands in your terminal to clone the repository, install dependencies, and run the project locally:

# Clone the repository
git https://huggingface.co/amirsoahil101/Bank_Telemarketing_Success_Prediction_using_ANN

# Navigate into the project directory
cd YOUR_PROJECT_FOLDER_NAME

# Install required packages
pip install -r requirements.txt

# Run the Streamlit application
streamlit run app.py