Enhancing Logistic Regression for Predicting Habitat Suitability of Scottish Crossbills (Loxia Scotica)

The prediction of habitat suitability for migratory avian species in response to climate change is a critical challenge in ecological conservation. This study focuses on the Scottish Crossbill (Loxia scotica), employing an enhanced Logistic Regression algorithm to address the limitations of traditional approaches. A significant issue with the existing algorithm was the use of default hyperparameters, including L2 regularization, a Limited-memory Broyden–Fletcher–Goldfarb–Shanno (BFGS) solver, and limited iterations. These settings constrained the algorithm’s generalizability and ability to adapt to the complexity of habitat prediction data. To overcome these challenges, the researchers utilized the PyCaret package, which facilitates comprehensive hyperparameter tuning by systematically exploring combinations of key parameters such as regularization strength and solver types, alongside cross-validation for robust performance evaluation. The integration of PyCaret significantly improved the algorithm’s performance. Compared to the existing Logistic Regression algorithm, the enhanced algorithm exhibited an 11.4% increase in accuracy, a 4.74% rise in the Area Under the Curve (AUC), and a 9% improvement in the F1-score during Test Set Evaluation. Specifically, the enhanced algorithm achieved an accuracy of 88%, an AUC of 88.99%, and an F1-score of 88%. These results highlighted the enhanced algorithm’s predictive capabilities and its robustness in identifying suitable habitats. The enhanced algorithm’s ability to predict habitat suitability more effectively underpins its potential for aiding conservation planning. By implementing systematic hyperparameter tuning, the enhanced algorithm not only achieves higher prediction accuracy but also minimizes bias and variance, paving the way for more reliable predictions of habitat suitability under changing climate conditions.