The random forest analysis predicted the inhabitation status of islands based on their microbial communities with high accuracy for free-living (91.7%), particle-attached (93.8%), and sediment (93.8%) models. The models had low out-of-bag errors (<0.083) and significantly outperformed random predictions (p<0.001). Variable importance was determined by a mean decrease in accuracy and GINI of at least 0.0020, 0.0030, and 0.0016 respectively.
The random forest analysis predicted the inhabitation status of islands based on their microbial communities with high accuracy for free-living (91.7%), particle-attached (93.8%), and sediment (93.8%) models. The models had low out-of-bag errors (<0.083) and significantly outperformed random predictions (p<0.001). Variable importance was determined by a mean decrease in accuracy and GINI of at least 0.0020, 0.0030, and 0.0016 respectively.
The random forest analysis predicted the inhabitation status of islands based on their microbial communities with high accuracy for free-living (91.7%), particle-attached (93.8%), and sediment (93.8%) models. The models had low out-of-bag errors (<0.083) and significantly outperformed random predictions (p<0.001). Variable importance was determined by a mean decrease in accuracy and GINI of at least 0.0020, 0.0030, and 0.0016 respectively.