"Loess Regression is the most common method used to smoothen a volatile time series. It is a non-parametric methods where least squares regression is performed in localised subsets, which makes it a suitable candidate for smoothing any numerical vector."

Loess regression can be done on a numerical vector to smoothen it. The predictor variable can just be indices from 1 to number of observations. If other predictors are available, feel free to test them out too. data(fuel, package="fpp") # initialize data
loessMod10 <- loess(Cost ~ c(1:nrow(fuel)), data=fuel, span=0.10) # span arg control the degree of smoothing
smoothed10 <- predict(loessMod10) # get smoothed output
predicted <- predict(loessMod10, newData) # predict on new data