Weighted time series forecasting¶

Although historical data is available in many real use cases of forecasting, not all are reliable. Some examples of these scenarios are:

• IoT sensors: within the Internet of Things, sensors capture the raw data from the physical world. Often the sensors are deployed or installed in harsh environments. This inevitably means that the sensors are prone to failure, malfunction, and rapid attrition, causing the sensor to produce unusual and erroneous readings.

• Factory shutdown: every certain period of operation, factories need to be shut down for repair, overhaul, or maintenance activities. These events cause production to stop, generating a gap in the data.

• Pandemic (Covid-19): the Covid 19 pandemic changed population behavior significantly, directly impacting many time series such as production, sales, and transportation.

The presence of unreliable or unrepresentative values in the data history is a major problem, as it hinders model learning. For most forecasting algorithms, removing that part of the data is not an option because they require the time series to be complete. An alternative solution is to reduce the weight of the affected observations during model training. This document shows two examples of how skforecast makes it easy to apply this strategy.