Forest fires are a growing global concern, causing environmental damage, threatening biodiversity, and endangering human lives. In Algeria, the Bejaia and Sidi-Bel Abbes regions face heightened risk due to rising temperatures, dry conditions, and strong winds. Predictive models can help forecast fire risks based on environmental factors, enabling early intervention strategies.
This blog explores the use of linear regression to predict the Fire Weather Index (FWI) and logistic regression to predict fire occurrences. Using the Algerian Forest Fires Dataset, we analyze how temperature and wind speed influence fire risks and occurrences.
Dataset Overview
The Algerian Forest Fires Dataset includes data from two regions: Bejaia and Sidi-Bel Abbes. Key variables include:
- Temperature
- Relative Humidity (RH)
- Wind Speed (Ws)
- Fire Weather Index (FWI)
- Fire occurrence class ("fire" or "not fire")
The data was preprocessed, with "fire" encoded as 1 and "not fire" as 0 for logistic regression. Linear regression focused on modeling the relationship between temperature and FWI.
Methodology
Linear Regression
Linear regression models the relationship between a dependent variable (FWI) and an independent variable (temperature). The analysis was conducted separately for each region, splitting the data into 70% training and 30% testing sets to evaluate the model’s performance on unseen data.
Logistic Regression
Logistic regression is used for binary classification problems, predicting the likelihood of fire occurrence based on wind speed. This analysis was also region-specific, with similar training and testing data splits.
Results
Linear Regression Findings
- In the Bejaia Region, the model achieved a Mean Squared Error (MSE) of 32.44, with a regression coefficient of 1.25, indicating that FWI increased by 1.25 units per 1°C rise in temperature.
- In the Sidi-Bel Abbes Region, the MSE was 35.68, with a regression coefficient of 1.12, suggesting a weaker but still significant relationship between temperature and FWI.
Takeaway: Temperature is a significant predictor of fire risk, but regional variations affect the strength of this relationship.
Logistic Regression Findings
- In the Bejaia Region, the model had an accuracy of 48.6%, with 19 false positives and no false negatives, suggesting wind speed alone is insufficient for predicting fire occurrences.
- In the Sidi-Bel Abbes Region, accuracy improved to 59.5%, but the model consistently overpredicted fires, indicating a bias toward fire occurrences.
Takeaway: Wind speed is an important factor, but additional variables like humidity or vegetation dryness are needed for accurate predictions.
Discussion
The findings underscore the potential of regression models for predicting fire risk while highlighting their limitations. Linear regression showed strong predictive power for temperature’s influence on FWI, particularly in the Bejaia Region. However, logistic regression struggled to accurately predict fire occurrences based solely on wind speed, emphasizing the multifactorial nature of fire dynamics.
One limitation of the study is its reliance on univariate models. Fire risk is influenced by multiple environmental factors, including rainfall, humidity, and vegetation type. Future models should adopt multivariate approaches to capture these complex interactions more effectively.
Conclusion
Linear regression demonstrated the importance of temperature in predicting fire risk, while logistic regression revealed the inadequacy of wind speed as a sole predictor for fire occurrences. To improve prediction accuracy, future models should integrate additional variables and address class imbalance issues in the dataset.
By refining these models, fire management authorities can develop more effective strategies for wildfire prevention, resource allocation, and early intervention. Predictive analytics holds great promise in mitigating the devastating impacts of forest fires, especially in vulnerable regions like Algeria.