Earthquake Magnitude Prediction Model

🔹 Problem Statement

Predict the magnitude of earthquakes using historical earthquake event data.
Accurate magnitude prediction helps assess potential risks, improve disaster preparedness, and support seismological research.

---

Dataset Description

Source: Kaggle – Earthquake Dataset
Target Variable: magnitude
Features used:

• longitude, latitude – location of the earthquake
• year, month – time-based features
• depth, sig, nst, dmin, rms, gap, tsunami – seismological attributes
• magType, type – event type indicators

---

Exploratory Data Analysis (EDA)

Performed detailed analysis to understand relationships and data trends:

• Distribution of Magnitudes: Most earthquakes have magnitudes between 3.0 and 6.0.
• Depth vs Magnitude Correlation: Deeper earthquakes show slightly lower magnitudes on average.
• Regional Patterns: Higher activity clusters observed near specific latitude-longitude ranges.
• Temporal Trends: Magnitude patterns analyzed over time using year and month.

Key visualizations include:

• Magnitude distribution histogram
• Scatter plot of depth vs magnitude
• Heatmap of feature correlations
• Map of earthquake locations (longitude vs latitude)

---

Machine Learning Models

Model	RMSE	MAE	R² Score
Gradient Boosting	0.233	0.0299	0.9184
Random Forest	0.238	0.0277	0.9150
Ridge Regression	0.268	0.130	0.8920
Linear Regression	0.268	0.130	0.8920
Lasso Regression	0.272	0.131	0.8889

Best Model: Gradient Boosting Regressor (R² = 0.918)

It achieved the highest R² score (0.92) and lowest RMSE (0.23), showing strong predictive accuracy.

---

Feature Importance (from Gradient Boosting)

Top influential features for magnitude prediction:

1. sig (significance factor)
2. dmin (distance to nearest station)
3. tsunami
4. rms (root mean square of travel time residuals)
5. magType

Visualized below:

• Actual vs Predicted Magnitude (Scatter Plot)
• Residual Distribution (Prediction Errors)
• Feature Importance (Gradient Boosting)

---

Model Insights

• The sig feature dominates prediction importance, indicating event significance is a strong indicator of magnitude.
• Geographical coordinates (latitude, longitude) contribute less, suggesting local variations are less dominant than seismic parameters.
• Ensemble models (Random Forest, Gradient Boosting) outperform linear models due to their ability to capture nonlinear relationships.

---

Evaluation: Actual vs Predicted Magnitude

The actual vs predicted plot shows strong alignment:

• Points are closely scattered around the diagonal line (ideal prediction).
• Gradient Boosting demonstrates minimal deviation and low residual error.

---

Results & Insights

• Gradient Boosting performed best with lowest RMSE (0.23) and highest R² (0.92).
• Geographic location (longitude, latitude) and seismic station parameters (nst, dmin) were most influential.
• Linear models underperformed compared to ensemble methods.

Technologies Used

• Language: Python 3.x
• Libraries:
• pandas, numpy
• matplotlib, seaborn
• scikit-learn

How to Run

1. Clone this repository:

   git clone https://github.com/<your-username>/earthquake-prediction.git

2. Run the Jupyter Notebook jupyter notebook earthquake_magnitude_predictions.ipynb