Time Series Forecast Benchmark

Benchmarking app for time series forecasting methods.

---

Usage

App currently deployed and runnable at: https://domdatascience.shinyapps.io/time-series-forecast-benchmark/

App Features

Compare different time series forecasting methods using a single plot:

• The user can browse the file system and input any file having a monthly time series;
• A date range can be defined for the input time series with start and end date. This range is used to form the "Time" x-axis of the plot;
• The user can also input any number of months to be forecasted using 10 forecasting models:
  1. Naive Forecasting;
  2. Mean Forecast;
  3. Random Walk Forecast with Drift;
  4. Holt's Linear;
  5. Holt-Winters Additive;
  6. Holt-Winters Multiplicative;
  7. Holt-Winters Multiplicative with Drift;
  8. ARIMA;
  9. Linear Model;
  10. Neural Network.
• A plot is created with the original time series and the amount of months the user wants to forecast;
• The user can select models to make their forecasting curves visible on the plot;
• A small table is created for each one of the 10 forecasting models. Error metrics and other quality measures for each prediction method are shown.

User Interface Sample

Shiny is a framework that allows users to develop web apps using R and embedded web languages, such as CSS and HTML. Shiny apps focus on objectiveness and simplicity: only one or two R scripts have all the code for the app.

This app development started with knowledge and tools discussed during the course "Data Science Bootcamp" by Fernando Amaral. The app has been upgraded and personalized, adding new functionalities.