Which techniques are helpful to forecast and classify time-series data, and what are the benchmark models to compare with?
Hi, Exponential smoothing methods, neural network, evolutionary algorithms are experimented by researchers.
Do you mean to classify in different groups, thus clustering?
Look at "Comparing Time-Series Clustering Algorithms in R Using the dtwclust Package". Dynamic Time Warping is the most usual way. Else you need to look at them from the phase-space.
Forecast depends on the length of the series and properties.
1. If short, then ARIMA or ES are classical methods.
2. If the time series have a trend, then you need ARIMA or ES
3. For long time series or with long seasonality, use Machine Learning methods: ANN, RF, Boosting Tree.
time series focasting can also be classified as univariate and multivariate forecasting.
In univariate forecasting, most widely used models are ARIMA, ARMA etc.,
As for trend and seasonal time series, I think that you can use k-NN search
for time series forecasting.
for classification, an old paper by me, paterlini and minerva may help. FP
https://towardsdatascience.com/attention-for-time-series-classification-and-forecasting-261723e0006d
- Badges
- Science topic
- Similar topics
- Control Systems Engineering
- Filtering