BASTION: Bayesian Adaptive Seasonality and Trend Decomposition Incorporating Outliers and Noise

BASTION is an R package providing a Bayesian framework for decomposing univariate time series into meaningful components. It offers several key advantages over existing decomposition methods:

• Locally adaptive estimation of trend and seasonality.
• Explicit modeling of outliers and time-varying volatility.
• Robust uncertainty quantification through credible intervals for all components.

---

Installation

# Install devtools package if not already installed
install.packages("devtools")
devtools::install_github("Jasoncho0914/BASTION")

---

Quick Start

The main function is fit_BASTION . Following is a quick demonstration using the Airline Passenger data between 1949 to 1960.

data(airtraffic)
fit <- fit_BASTION(y = airtraffic$Int_Pax,
                   Ks = list(12))
#Trend
plot(fit$summary$p_means$y,ylim = c(-1,13),ylab = "Trend")
lines(fit$summary$p_means$Trend,lwd = 4)

#Yearly Seasonality
plot(fit$summary$p_means$Seasonal12,type= "l",lwd = 4,ylab = "Seasonality (k=12)")

#Remainder
plot(fit$summary$p_means$Remainder,type = "l",lwd = 4,ylab = "Remainder")

---

Learn More

• Overview: Key features and practical applications of BASTION are summarized.
• Preprint: For an in-depth discussion, see this accompanying paper.

---

Citation

@Manual{BASTION,
  title        = {BASTION: Bayesian Adaptive Seasonality and Trend Decomposition Incorporating Outliers and Noise},
  author       = {Jason B. Cho},
  year         = {2025},
  url          = {https://github.com/Jasoncho0914/BASTION}
}