Time Series Forecasting with Shiny

Launch the Shiny app!

This Shiny app provides an interactive user interface to visualise and forecast time series.

Users can upload their own CSV with single or multiple daily time series. The user interace allows users to compare fitted time series models and forecasts with several algorithms including:

• line of best fit (regression)
• moving average
• exponential smoothing (simple & Holt-Winters)
• ARIMA with Fourier Transform
• TBATS
• hybrid forecast ensemble model

Packages Utilised

• Many of the models draw on the excellent forecast package from Rob J Hyndman
• Visualisations are interactive using Dygraphs for R
• and of course Shiny. Excellent.

Features

• interactive time series visualisations showing series, fitted & forecast
• forecast between 1 and 120 periods with 95% confidence intervals
• collect & compare forecasts using various models with the same parameters - see the compare tab. Note: changing parameters refreshes the compare view
• view and download forecasts and 95% confidence interval predictions
• include/exclude weekends
• adjust relevant models for shocks using exogenous dummy variables. Note: the current file used is a CSV of Australian public holidays for demo purposes only.

File Format

Note: there are currently no file format validations built in. Files should have format as per the table below. An example is provided here.

Date	Series_1	Series_n (optional)
dd/mm/yyyy	series_1_value_1	series_n_value_1
dd/mm/yyyy	series_1_value_2	series_n_value_2
dd/mm/yyyy	series_1_value_n	series_n_value_n

Known Issues

This is very much alpha with a number of issues to fix and features to add including:

• file format validations
• extend model to other series eg. minute, hourly, weekly etc
• handling for date formats other than dd/mm/yyyy
• add conditional modelling parameters eg. moving average alpha and periods
• ability to upload own exogenous variables
• ability to define series frequency - currently hardcoded to 365.25/7 (weekly)