This project focuses on analyzing a forest fire dataset through exploratory data analysis, feature engineering, and predictive machine learning modeling using both linear regression and random forest algorithms. The dataset includes details about forest fires such as spatial coordinates, month, day, weather conditions, and the area burned.
Key aspects of the project include:
Data Preprocessing: Removing duplicates, converting month and day into numerical values, and applying a log transformation to normalize the skewed burned area feature.
Visualization: Using various plots to explore fire occurrences across different months and days, as well as examining the relationships between weather variables and the area burned.
Predictive Modeling: Developing and evaluating linear regression and random forest models to predict the log-transformed burned area. The models' performance is evaluated using RMSE, and feature importance is visualized from the random forest model. The project aims to understand the factors contributing to forest fires and improve the accuracy of burned area predictions.
Summary of Results: Linear Regression RMSE: 1.670814 Random Forest RMSE: 0.4254481
Make sure to check the ForestFires.Names file
