A Machine Learning Algorithm-Based Approach (MaxEnt) for Predicting Habitat Suitability of Formica rufa

Abstract

Machine learning techniques are quite effective for simulating species habitat appropriateness. Species distribution models are statistical algorithms founded on the ecological niche idea. These models estimate the association between existing species records and the environmental and spatial characteristics of the habitat. From 2022 to 2023, a field survey was conducted in the Kastamonu Forest Enterprise, resulting in the identification of 267 active Formica rufa nests. The habitat preferences of F.rufa were assessed based on factors such as stand characteristics, topography and climatic variables. MaxEnt, a prevalent machine learning technique for predicting species habitat suitability, was employed in the habitat suitability modelling of F. rufa. 30 distinct variables were employed in the modelling process. Receiver Operating Characteristic (ROC) analysis examined model accuracy. AUC was 0.941 for training data and 0.946 for test data. With 39.5% of the model, the development stage is the most important variable for F. rufa habitat selection. The development stage, productivity and temperature annual range (BIO7) variables make up 75.1% of the model. The habitat suitability map shows that 79% of F. rufa nests are in moderately and highly appropriate areas. The F. rufa group, widely prevalent in northern hemisphere forests, significantly impacts forest ecosystems and is recognised as the foremost bioindicator species within these environments. Determining the elements that affect habitat selection by these species is essential for their conservation and management.