Thesis: Sinkhole Susceptibility Analysis Using Machine Learning

Overview

This project involves analyzing sinkhole susceptibility using a variety of machine learning algorithms. The dataset contains multiple geological and environmental features, which are used to predict the occurrence of sinkholes. The analysis includes data preprocessing, feature selection, model building, and evaluation of different machine learning models. In the article, you can mention the following:

1. Pre-processing of the Dataset: The initial pre-processing of the dataset was conducted using ArcGIS, ensuring that the data was accurately prepared for further analysis.

2. Exporting the RF Model to ArcGIS: After training the Random Forest (RF) model, it was exported to ArcGIS to generate a detailed sinkhole susceptibility map for the study area, allowing for a seamless integration of the predictive model with advanced geospatial visualization tools.

Requirements

Ensure you have the following Python libraries installed:

• numpy
• pandas
• matplotlib
• scikit-learn

You can install them using pip:

pip install numpy pandas matplotlib scikit-learn

Data

The dataset sinkhole_data.csv contains the following columns:

• ID: Unique identifier for each observation.
• Sinkhole: Target variable (1 indicates sinkhole occurrence, 0 indicates no sinkhole).
• Land use: Land use type.
• Substrate: Type of substrate.
• DTS: Distance to streams.
• DTM: Distance to mines.
• HD: Hydraulic head difference.
• Depth to Bedrock: Depth to bedrock.
• DTF: Distance to fault.
• DWTSA: Depth to water table of the surficial aquifer.
• DTD: Distance to depressions.

Steps

1. Importing Required Libraries

import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.feature_selection import mutual_info_classif
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.decomposition import PCA
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.neural_network import MLPClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import confusion_matrix, accuracy_score, roc_curve, roc_auc_score, precision_recall_fscore_support, precision_recall_curve

2. Loading and Understanding the Data

df = pd.read_csv('C:/Users/MUILI OLANREWAJU/Research Thesis/Data/Processed data/sinkhole_data.csv')
print(df.head())
print(df.columns)
print(df.shape)
print(df.info())
print(df.describe())
print(df.isna().sum())

3. Preprocessing the Data

• Slice relevant columns.
• Re-order columns to have non-categorical variables first.
• Perform one-hot encoding for categorical variables.

df = df[df.columns[1:]]
categorical_variables = ['Substrate', 'Land use', 'Sinkhole']
non_categorical_variables = list(set(df.columns) - set(categorical_variables))

order = non_categorical_variables + categorical_variables
df = df[order]

4. Feature Scaling

Standardize the dataset before feeding it into machine learning models.

scaler = StandardScaler()
X = scaler.fit_transform(df.iloc[:, :-1].values)
y = df['Sinkhole'].values

5. Principal Component Analysis (PCA)

Visualize the dataset in two dimensions using PCA.

pca = PCA(n_components=2, random_state=0)
principalComponents = pca.fit_transform(X)
principalDf = pd.DataFrame(data=principalComponents, columns=['PC1', 'PC2'])

6. Model Building

Split the dataset into training and testing sets.

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0)

7. Applying Machine Learning Algorithms

Several machine learning algorithms are applied with hyperparameter tuning using GridSearchCV:

• Random Forest
• K Nearest Neighbors
• Logistic Regression
• Support Vector Machine
• Multilayer Perceptron Neural Network

Example for Random Forest:

param_grid = {
    'n_estimators': [200, 300, 400, 500],
    'max_features': ['auto', 'sqrt', 'log2'],
    'max_depth': np.arange(10, 30, 2),
    'criterion': ['gini', 'entropy']
}
grid = GridSearchCV(estimator=RandomForestClassifier(random_state=0), param_grid=param_grid)
grid.fit(X_train, y_train)
print('Best score:', grid.best_score_)
print('Best hyperparameters:', grid.best_params_)

8. Model Evaluation

Evaluate the models using confusion matrix, accuracy, ROC AUC, precision, recall, and F1 score.

rf_conf_mat = confusion_matrix(y_test, rf_pred)
rf_acc = accuracy_score(y_test, rf_pred)
rf_roc_auc = roc_auc_score(y_test, rf_proba[:, 1])

print('Accuracy:', rf_acc)
print('ROC AUC:', rf_roc_auc)

9. Save Results and Figures

Save PCA plots and model performance results for further analysis.

plt.savefig(os.path.join('..data\\figures', 'pca.png'), dpi=300)

Conclusion

The project involves detailed analysis using various machine learning techniques to predict sinkhole susceptibility. The optimal model and its performance metrics provide insights into which features contribute the most to predicting sinkhole occurrences.

Author

Olanrewaju Muili

License

This project is licensed under the MIT License.

Acknowledgments

Thanks to The Doe Run Company for providing the dataset and support for this project.

Data

Sinkholes http://publicfiles.dep.state.fl.us/otis/gis/data/FGS_SUBSIDENCE_INCIDENTS.zip

Streams https://geodata.dep.state.fl.us/search?q=geology&sort=-modified

Depressions http://publicfiles.dep.state.fl.us/otis/gis/data/LAND_SURFACE_ELEVATION_24.zip

Active mines https://floridadep.gov/fgs

Fault https://geodata.dep.state.fl.us/search?q=geology&sort=-modified

Substrate http://publicfiles.dep.state.fl.us/OTIS/GIS/data/GEOLOGY_STRATIGRAPHY.zip

Depth to bedrock https://geodata.dep.state.fl.us/search?q=geology&sort=-modified

Depth to water table of Surficial aquifer https://geodata.dep.state.fl.us/search?q=geology&sort=-modified

Head difference https://geodata.dep.state.fl.us/search?q=geology&sort=-modified

Land use http://publicfiles.dep.state.fl.us/otis/gis/data/STATEWIDE_LANDUSE.zip

Literature

https://pdfs.semanticscholar.org/608b/627afb47103d4456ea64abf5a9f74dbec1f8.pdf

https://www.sciencedirect.com/science/article/abs/pii/S0022169420305096?via%3Dihub

https://www.mdpi.com/2076-3417/10/15/5047

https://www.nature.com/articles/s41598-019-43705-6

https://onlinelibrary.wiley.com/doi/10.1002/ldr.3255