classoptimr: classoptimr: A package for identifying optimal classification...

In AndreasChristianHill/classoptimr: Optimal classification schemes for prediction models with continuous response variable

Description

Core of this package is a heuristic optimization procedure (Simulated Annealing) that allows for identifying optimal classification schemes for models that use continuous response variables and produce predictions on a continuous scale. The implemented methods were primarily developed to quantify the classification accuracy of prediction maps based on statistical models that provide predictions on a continuous scale (see references). In many cases, these continuous predictions are afterwards discretized into classes for better visualization purposes without considering the resulting accuracies of the created classification scheme. In a more general modelling context, the optimization method can also be used to detect non-constant prediction performance of statistical models.

Functions

The package provides three main functions to apply:

• HSMclass Function to identify an optimal classification scheme for a predefined number of classes GIVEN a set of response reference and corresponding predicted values.

• classAccuracy Function to evaluate a classification scheme by calculating various classification accuracy measures.

• create_qml Function to create a qml-file of a classification scheme for visualization in the open source Geographical Information System QGIS.

References

Hill, A., Breschan, J., & Mandallaz, D. (2014). Accuracy assessment of timber volume maps using forest inventory data and LiDAR canopy height models. Forests, 5(9), 2253-2275.

AndreasChristianHill/classoptimr documentation built on May 29, 2019, 12:23 p.m.