ANN Projects

Welcome to my Artificial Neural Network (ANN) repository!
This repository features end-to-end Deep Learning projects built with TensorFlow and Keras, demonstrating how ANNs can solve both classification and regression problems in real-world data scenarios.

Each project includes full data preprocessing, model building, tuning, and Streamlit deployment, making it both educational and production-ready.

---

Projects Overview

1. Spotify Churn Prediction – Classification

An end-to-end Artificial Neural Network (ANN) project that predicts whether a Spotify user is likely to churn based on their usage patterns and listening behavior.
The system helps identify at-risk users, enabling proactive retention strategies.

Framework: TensorFlow / Keras
Model Type: Binary Classification
Deployment: Streamlit (Live App)

Key Features

• Built an ANN with multiple hidden layers using ELU activations.
• Implemented StandardScaler and ColumnTransformer for preprocessing.
• Tuned model hyperparameters via GridSearchCV (KerasClassifier wrapper).
• Achieved ~74.8% validation accuracy after tuning.
• Real-time predictions through a Streamlit interface.

Technologies Used
Python | TensorFlow/Keras | scikit-learn | pandas | numpy | seaborn | matplotlib | Streamlit

Live App: View on Streamlit

---

2. Beats Per Minute (BPM) Prediction – Regression

A Deep Learning regression model that predicts the Beats Per Minute (BPM) of tracks using an ANN built with TensorFlow/Keras, developed for the Kaggle Playground Series - S5E9.

** Framework:** TensorFlow / Keras
** Model Type:** Regression
** Tuning:** RandomizedSearchCV with KerasRegressor

Key Features

• Full data exploration and preprocessing using pandas, matplotlib, seaborn.
• Scaled features using StandardScaler for stable ANN convergence.
• Applied EarlyStopping (patience = 35) to prevent overfitting.
• Tuned model parameters such as activations, optimizers, and epochs.
• Optimized model performance using RMSE as the evaluation metric.

Technologies Used
Python | TensorFlow/Keras | scikit-learn | pandas | numpy | seaborn | matplotlib

Competition Source: Kaggle Playground S5E9

---

Common Tech Stack

Category	Tools / Libraries
Language	Python 3.10
Deep Learning	TensorFlow, Keras
Data Handling	pandas, numpy
Preprocessing	StandardScaler, OneHotEncoder, ColumnTransformer
Model Tuning	scikeras (KerasClassifier/KerasRegressor), GridSearchCV, RandomizedSearchCV
Visualization	matplotlib, seaborn
Deployment	Streamlit
Utilities	pickle, dotenv

---

Key Learnings & Highlights

• Designed ANN architectures for both regression and classification.
• Applied systematic hyperparameter optimization for deep learning models.
• Leveraged modern deployment practices for live AI apps using Streamlit.
• Developed scalable preprocessing pipelines with scikit-learn transformers.
• Delivered interpretable results for business and creative applications alike.