Global Earthquake–Tsunami Risk Prediction using Machine Learning

🔍 Overview

End-to-end AI/ML project that predicts tsunami potential from global earthquake data (2001–2022).

Built using Python, Scikit-Learn, and Streamlit, this project explores how machine learning can analyze seismic data and forecast tsunami risk.

> 📊 Dataset: 782 global earthquakes (2001–2022) with 13 features and binary tsunami labels. Global Earthquake-Tsunami Risk Assessment Dataset | Kaggle

> 💡 Key Goal: Classify earthquake events as “Tsunami” or “Non-Tsunami.”

⚙️ Workflow

1. Data Preprocessing & EDA

• Cleaned 782 records, engineered features (mag_depth_ratio, depth_category)
• Visualized global distribution, magnitude–depth relationships, and correlations

2. Modeling & Evaluation

• Compared Logistic Regression (AUC 0.93) vs Random Forest (AUC 0.965)
• Tuned hyperparameters using GridSearchCV → best AUC 0.966

3. Model Variants

• tsunami_rf_model.pkl → includes temporal features (Year, Month)
• tsunami_rf_noyear.pkl → excludes temporal bias, for future prediction

Reason:
Year-based features improve accuracy for 2001–2022,
but make the model biased toward specific years.
Therefore, a second “future-ready” model was built without Year/Month

4. Deployment

• Built an interactive web dashboard using Streamlit
• Mode 1: Historical analysis (maps, trends)
• Mode 2: Real-time prediction (user inputs magnitude, depth, etc.)

🧠 Key Insights

• Shallow, high-magnitude quakes near coastlines have highest tsunami potential
• Temporal features improved accuracy historically, but reduced generalization
• Removing time-based features created a future-ready model

🖼️ Visual Highlights

🚀 Tech Stack

• Python, Pandas, NumPy
• Scikit-Learn, Matplotlib, Plotly
• Streamlit (interactive dashboard)
• Google Colab (training environment)

🧩 Results

• Best Model: Random Forest (Tuned)
• ROC–AUC: 0.966
• Balanced dataset (39% tsunami events)
• Fully deployed on Streamlit Cloud
• Interactive visualization + real-time tsunami prediction

🧱 Project Files

• /data/earthquake_data_tsunami.csv
• /models/tsunami_rf_model.pkl
• /models/tsunami_rf_noyear.pkl
• /app.py
• /requirements.txt

🖼️ Streamlit App Features

📊 1. Historical Analysis Mode

Menampilkan analisis data historis dan perilaku model:

• Global scatter map (Tsunami vs Non-Tsunami)
• Magnitude vs Depth scatter
• Heatmap zona risiko tinggi
• Yearly trends dan frekuensi
• Magnitude distribution
• Correlation heatmap

🔮 2. Prediction Mode

• Memprediksi potensi tsunami dari parameter gempa baru:
• Input: magnitude, depth, latitude, longitude, dll
• Menghitung fitur turunan: depth_category, mag_depth_ratio
• Output:
  • Tsunami Probability (%)
  • Prediction Result (Likely / Not Likely)
  • ⁠Epicenter Map Visualization

🚀 Key Results

• Final Model ROC-AUC: 0.966
• Balanced dataset (39% tsunami events)
• High prediction reliability
• Fully deployed on Streamlit for interactive visualization

🧭 Tech Stack

🧩 Project Structure

project/
│
⁠├── data/
⁠│  └── earthquake_data_tsunami.csv
│
├── notebooks/
│  ├── 01_data_preprocessing.ipynb
│  ├── 02_model_training.ipynb
│  └── 03_model_evaluation.ipynb
│
├── models/
│  ├── tsunami_rf_model.pkl
│  └── tsunami_rf_noyear.pkl
│
├── app.py
├── requirements.txt
└── README.md

🧠 Insights & Learnings

• Temporal features (Year) dapat menyebabkan bias prediksi — dihapus untuk generalisasi.
• Gempa dangkal dengan magnitude besar di dekat pantai paling berpotensi tsunami.
• ⁠Kombinasi parameter fisik dan spasial meningkatkan akurasi model.
• ⁠Streamlit mempermudah penyajian hasil machine learning secara interaktif dan intuitif.

🔗 Live Demo: Demo

📂 Source Code:Github