NMSE: NMSE - Normalized Mean Squared Error

In jdestefani/UEMTS: Univariate Error Measures for Time Series forecasting

Description

NMSE = 1/n (∑_{t=0}^{n} (y_t-y_hat_t)^2)/(var(y_t))

Usage

NMSE(y, y_hat, normalizing_variance = NULL, result_type = c("measure", "raw"))

Arguments

y	True values of the time series
y_hat	Forecasted values of the time series
normalizing_variance	Variance of the training set
result_type	String indicating whether the function should return the raw values ("raw"), or the computed measure ("measure")

Value

According to the value of result_type:

• measure -> NMSE computed according to the formula

• raw -> Vector containing one NMSE for each time step

Examples

y <- AirPassengers
splitting_point <- round(2*length(y)/3)
y_train <- y[1:splitting_point]
h <- 5

y_test <- y[(splitting_point+1):(splitting_point+h)]
y_hat_naive <- rep(tail(y_train,1),h)
results <- NMSE(y_test,y_hat_naive)

jdestefani/UEMTS documentation built on Dec. 20, 2021, 10:05 p.m.