🦀 Crab Age Prediction Using Regression Models

📘 Overview

This project predicts the age of crabs using their physical measurements such as length, weight, diameter, and shell weight. The primary goal is to explore multiple regression algorithms, engineer meaningful features, and identify the model that best estimates crab age.

The project demonstrates:

• End-to-end machine learning workflow
• Advanced feature engineering
• Multiple pipeline architectures
• Model comparison and fine-tuning with GridSearchCV
• Evaluation using Root Mean Squared Error (RMSE)

---

📊 Dataset

• Source: Crab Age Prediction Dataset (Kaggle)
• Description: The dataset includes various biological and physical measurements of crabs. Each record corresponds to a crab, and the target variable is its age (in years).

Features:

Feature	Description
Sex	Male, Female, or Infant
Length	Largest distance from the front to the back of the shell (mm)
Diameter	Widest point perpendicular to the length (mm)
Height	Height of the crab (mm)
Weight	Whole weight of the crab (g)
Shucked Weight	Weight of meat only (g)
Viscera Weight	Weight of gut content (g)
Shell Weight	Weight of shell after being dried (g)
Target: Age	Estimated age of the crab (years)

---

⚙️ Project Workflow

1. Data Preprocessing

• Handled missing values and outliers

• Encoded categorical variables (Sex) using OneHotEncoder

• Scaled numerical features with multiple strategies:

  • StandardScaler
  • RobustScaler
  • MinMaxScaler

• Built three custom preprocessing pipelines to test scaling and imputation variations

2. Feature Engineering

Custom features were added to enrich the dataset:

• Volume = Length × Weight × Height
• Shell Weight Ratio = Shell Weight / Weight

3. Model Training

Trained and compared multiple regression algorithms:

• Linear Regression
• Decision Tree Regressor
• Random Forest Regressor
• Gradient Boosting Regressor
• XGBoost Regressor
• LightGBM Regressor (Best Performer)

4. Model Fine-Tuning

Fine-tuned the LightGBM model using GridSearchCV with parameters:

5. Evaluation

Used Root Mean Squared Error (RMSE) to evaluate model performance on training, validation, and test sets.

---

📈 Results

The LightGBM model provided the best generalization with minimal overfitting, making it the final chosen model.

Metric	Train RMSE	Validation RMSE	Test RMSE
Value	1.956	2.061	2.417

These results show that the model generalizes well across unseen data, with only a small increase in error from validation to test sets.

---

📊 Visualizations

• Feature Correlation Heatmap
• Distribution of Numerical Features
• Outlier Detection (Boxplots)
• Feature Importance (LightGBM)

---

🧠 Tech Stack

• Python 3.10+
• pandas, numpy, matplotlib, seaborn
• scikit-learn
• XGBoost
• LightGBM

---

🚀 How to Run

1. Clone the repository:

   git clone https://github.com/MenaElghaysha/Crab-Age-Prediction-Using-Regression-Models.git
   cd crab-age-prediction

2. Open the notebook:

   jupyter notebook "Crab Age Prediction.ipynb"

3. Run all cells to reproduce the results.

---

🧩 Future Improvements

• Add more biological features for better accuracy
• Implement cross-validation visualization
• Experiment with deep learning regression
• Automate pipeline tuning with Optuna or Bayesian Optimization

---

👤 Author

Mena H. Elghaysha Machine Learning Engineer & Data Analyst

---

📜 License

This project is licensed under the MIT License — feel free to use and modify it.