Through the Random Forest: Ontogeny as a study system to connect prediction to explanation
Simon, S.; Glaum, P.; Valdovinos, F. S.
Show abstract
As modeling tools and approaches become more advanced, ecological models are becoming more complex and must be investigated with novel methods of analysis. Machine learning approaches are a powerful toolset for exploring such complexity. While these approaches are powerful, results may suffer from well-known trade-offs between predictive and explanatory power. We employ an empirically rooted ontogenetically stage-structured consumer-resource model to investigate how machine learning can be used as a tool to root model analysis in mechanistic ecological principles. Applying random forest models to model output using simulation parameters as feature inputs, we extended established feature analysis into a simple graphical analysis. We used this graphical analysis to reduce model behavior to a linear function of three ecologically based mechanisms. From this model, we find that stability depends on the interaction between internal plant demographics that control the distribution of plant density across ontogenetic stages and the distribution of consumer pressure across ontogenetic stages. Predicted outcomes from these linear models rival accuracy achieved by our random forests, while explaining results as a function of ecological interactions.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Top predators govern multitrophic diversity effects in tritrophic food webs 94%
- Order matters: Autocorrelation of temperature dictates extinction risk in populations with nonlinear thermal performance 93%
- Slower, but deeper community change: anthropogenic impacts on species temporal turnover are regulated by intrinsic dynamics 92%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.