ARIMA
Alternatives
ETS
Exponential Smoothing
Model: y
t
=
t1
+ b
t1
+ s
tm
+ ϵ
t
Use: Data with
trend and seasonality
ü Search via PubMed
LSTM
Long Short-Term Memory networks
Model: h
t
= o
t
tanh(C
t
)
Use: Capturing long-
range dependencies
ü Search via PubMed
VAR
Vector Autoregression
Model: y
t
= A
1
y
t1
+ · · · + A
p
y
tp
+ u
t
Use: Multiple inter-
related time series
ü Search via PubMed
SVR
Support Vector Regression
Model: ˆy = w · ϕ(x) + b
Use: Time series regression tasks
ü Search via PubMed
Prophet
Developed by Facebook
Model: y(t) = g(t) + s(t) + h(t) + ϵ
t
Use: Handling seasonality and
holidays (cf. Neural Prophet)
ü Search via PubMed
XGBoost
Extreme Gradient Boosting
Model: Additive tree models
Use: Combined with feature
engineering for time series
ü Search via PubMed
TBATS
Exponential smooth-
ing state space model
Model: y
t
=
t
+ s
t
+ ϵ
t
Use: Complex seasonal patterns
ü Search via PubMed