A repository for organizing my research sources, notes, questions and code.
- Conceptual Understanding of Models: How do we statistically model presence only data using the MaxEnt model?
- Model Validation: How can we assess forecast skill in a quantitative manner; currently we use intuitive judgment?
- Model Selection: How can we compare performance between different MaxEnt models?
- Model Interpretation: How can we define the response variable that we are modeling? To what geographic range can we reliably extrapolate model?
- Model Simplification: MaxEnt uses complex transformations on covariates such as: linear, product, quadratic, hinge, threshold, categorical, are these necessary?
- Future Models: Do we need a collection of different models, could this improve performance? Do different models do better based on time of year? Could other modeling techniques be used?
- Human Encounter Likelihood Component: What are novel ways that we can include human behavioral patterns more explicitly in the model?
- Could using differnt numbers of background points influence forecast skill?
- Is the current method of selecting background points really effective in creating a null model for the desired geographic region?
- Determine if selected parameters are good and free from correlation.
- Usage of sampling bias grids, maxent supports a weighting matrix that could be developed to weight sampling points based on demographic knowledge of population density.
- Development of an ensemble forecast using BIOMOD package which features different macheine learning tools.
- Run model for each day using window of 20 days-> what is the right number of day January
A detailed description of the experiments and research methods are described in the Thesis Outline.
The code for this project is primarily written in R. It is currently private, and for those who have access the link to the analytics repository is below.