Goodness of Fit {ehaGoF}

Goodness of Fit

Description

Tests predicted and observed values for the goodness of fit with various criteria. The goodness of fit tests are used to test how well the model fits. Measures of goodness of fit typically summarize the argument between targets or observed values and the values expected or predicted under the model in question.

Usage

GoF(Observations, Predicts,
                nTermInAppr = 2,
                ndigit = 3,
                RMSE = TRUE,
                RRMSE = TRUE,
                SDR = TRUE,
                CV = TRUE,
                PC = TRUE,
                PI = TRUE,
                ME = TRUE,
                RAE = TRUE,
                MRAE = TRUE,
                MAPE = TRUE,
                MAD = TRUE,
                RSq = TRUE,
                ARSq = TRUE,
                AIC = TRUE,
                CAIC = TRUE)

Arguments

Observations

Observed values or target vector.

Predicts

Predicted values. Values produced or fitted by approximation or regression.

nTermInAppr

Number of terms used in approximation or regression model. Generally 2 for simple linear model. Default is 2.

ndigit

Number of digits in decimal places. Default is 3.

RMSE

Whether to show Root Mean Square Error statistics. Default is TRUE.

RRMSE

Whether to show Relative Root Mean Square Error statistics. Default is TRUE.

SDR

Whether to show Standard Deviation Ratio statistics. Default is TRUE.

CV

Whether to show Coefficient of Variance statistics. Default is TRUE.

PC

Whether to show Pearson's Correlation Coefficients statistics. Default is TRUE.

PI

Whether to show Performance Index statistics. Default is TRUE.

ME

Whether to show Mean Error statistics. Default is TRUE.

RAE

Whether to show Global Relative Approximation Error statistics. Default is TRUE.

MRAE

Whether to show Modified Relative Approximation Error statistics. Default is TRUE.

MAPE

Whether to show Mean Absolute Percentage Error statistics. Default is TRUE.

MAD

Whether to show Mean Absolute Deviation statistics. Default is TRUE.

RSq

Whether to show Coefficient of Determination (R-Squared) statistics. Default is TRUE.

ARSq

Whether to show Adjusted Coefficient of Determination (Adjusted R-Squared) statistics. Default is TRUE. Warning: nTermInAppr must be supplied.

AIC

Whether to show Akaike's Information Criterion statistics. Default is TRUE. Warning: nTermInAppr must be supplied.

CAIC

Whether to show Corrected Akaike's Information Criterion statistics. Default is TRUE. Warning: nTermInAppr must be supplied.

Author(s)

Prof. Dr. Ecevit Eyduran, TA. Alper Gulbe

References

Comparison of the Predictive Capabilities of Several Data Mining Algorithms and Multiple Linear Regression in the Prediction of Body Weight by Means of Body Measurements in the Indigenous Beetal Goat of Pakistan, Ecevit Eyduran, Daniel Zaborski, Abdul Waheed, Senol Celik, Koksal Karadas and Wilhelm Grzesiak, Pakistan J. Zool., vol. 49(1), pp 257-265, 2017.

Examples

# dummy inputs, independent variable
# integers from 0 to 9
inputs <- 0:9

# dummy targets/observed values, dependent variable
# a product of 2*times inputs minus 5 with some normal noise
targets <- -5 + inputs*2 + rnorm(10)

# linear regression model
model<-lm(targets~inputs)

# About the model
summary(model)

# Number of Terms
n = length(model$coefficients)

# model's predicted values against targets
predicted<-model$fitted.values

# using library ehaGoF for goodness of fit.
library(ehaGoF)

# Goodness of fit statistics
GoF(targets, predicted, nTermInAppr=n)