basaktamer/Abalone_Age_Prediction

🐌 Abalone Age Prediction (Regression)

This repository contains a full-stack data science solution for predicting the age of abalones using physical measurements. Developed for the Kaggle Playground Series (S4E4), the project transitions from raw data exploration to a deployed web application.

🚀 Live Demo

You can interact with the model here: [Insert your Hugging Face Space Link Here]

📊 Project Highlights

• —Model: Random Forest Regressor
• —Metric: Root Mean Squared Logarithmic Error (RMSLE)
• —Local Validation Score: 0.15303
• —Python Version: 3.10

🧠 Key Insights & Feature Engineering

• —Sex Mapping: Categorical variables were mapped manually ({'I': 0, 'M': 1, 'F': 2}) to preserve the biological distinction of Infants.
• —"Water Weight" Discovery: A custom feature was engineered to calculate the difference between the whole weight and the sum of shucked, viscera, and shell weights.
• —Noise Reduction: "Impossible" negative water weights (common in synthetic datasets) were clipped to zero, increasing feature correlation with the target from 0.428 to 0.463.
• —Target Optimization: Used a log1p transformation on the target variable (Rings) to align the training process with the RMSLE evaluation metric.

🛠 Tech Stack

• —Data: Pandas, NumPy
• —ML: Scikit-Learn, Joblib
• —App: Streamlit
• —Environment: Python 3.10 (VS Code Venv)

📁 Repository Structure

• —app.py: The Streamlit web application.
• —abalone_model.joblib: The serialized Random Forest model.
• —requirements.txt: Python dependencies.
• —README.md: Project documentation and Hugging Face metadata.