Time series forecasting to determine relative causal impact

A non-transitory computer readable storage medium tangibly embodying a computer readable program code having computer readable instructions that, when executed, causes a computer device to carry out a method of determining a forecast time series value, the method comprising:

• harvesting a first set of time series data of a first continuous time period comprising behavioral data;
• harvesting a second set of time series data of the first continuous time period comprising data of one or more external factors, acquired from a different source and based on a different type of data than the behavioral data;
• harvesting a continuation of the first set of the time series data comprising new behavioral data, for a second time period;
• harvesting a continuation of the second set of the time series data comprising new data of the one or more external factors for the second time period;
• calculating, based on a forecast algorithm executable on the processor, a forecast time series value of a future behavioral data as a continuation of the first set of the time series data for a third time period that is after the second time period;
• and
• identifying, based on the forecast algorithm, a relative causal impact between each external factor and the predicted time series value of the behavioral data, for the third time period, wherein:
• harvesting the behavioral data comprises applying a first training algorithm executable on the processor to the first set of the time series data to create a first statistical prediction model that does not include the one or more external factors;
• and
• harvesting the data of one or more external factors comprises applying an attention based deep learning algorithm executable on the processor to the second set of the time series data and the first statistical prediction model to create a second statistical prediction model;
• wherein both the first set and the second set of the time series data of the first and second statistical prediction models are of the same first continuous time period.

A method of determining a forecast time series value, the method comprising:

• harvesting a first set of time series data of a first continuous time period comprising behavioral data;
• harvesting a second set of time series data of the first continuous time period comprising data of one or more external factors acquired from a different source and based on a different type of data than the behavioral data;
• harvesting a continuation of the first set of the time series data comprising new behavioral data, for a second time period;
• harvesting a continuation of the second set of the time series data comprising new data of the one or more external factors, for the second time period;
• calculating, based on a forecast algorithm executable on the processor, a forecast time series value of a future behavioral data as a continuation of the first set of the time series for a third time period that is after the second time period;
• and
• identifying, based on the forecast algorithm, a relative causal impact between each external factor and the predicted time series value of the behavioral data, for the third time period, wherein:
• harvesting the behavioral data comprises applying a first training algorithm executable on the processor to the first set of the time series data to create a first statistical prediction model that does not include the one or more external factors;
• harvesting the data of one or more external factors comprises applying an attention based deep learning algorithm executable on the processor to the second set of the time series data and the first statistical prediction model to create a second statistical prediction model;
• and
• both the first set and the second set of the one or more time series of the first and second statistical prediction models are of the same first continuous time period.