Predictive analytics with forecasting model selection

摘要

Methods and apparatus for predicting network activity with forecasting model selection are disclosed. In one embodiment a network analytics forecasting model with a preferred model error measurement is selected from among a set of model error measurements for a set of network analytics forecasting models. The selecting includes, for each of a set of network analytics forecasting models, calculating a set of predicted network analytics metric values corresponding to individual ones of a set of actual network analytics metric values in a network analytics metrics data set, comparing individual ones of the set of predicted network analytics metric values to corresponding ones of the set of actual network analytics metric values, and generating a model error measurement reflecting the comparing. Future network analytics metric values are predicted using the network analytics forecasting model with the preferred model error measurement.

主权项

1. A method, comprising:
obtaining a network analytics metrics data set including a first set of raw network analytics metric values, the first set of raw network analytics metric values being associated with one or more marketing promotions; filtering the raw network analytics metric values to generate a reduced network analytics metrics data set, wherein the reduced network analytics metrics data set includes a second set of raw network analytics metric values; selecting, by one or more processors, a network analytics forecasting model from a set of network analytics forecasting models, wherein each of the set of network analytics forecasting models is associated with a corresponding model error measurement and a set of model parameters, wherein the selecting of the network analytics forecasting model comprises, for each of the set of network analytics forecasting models:
calculating a first set of predicted network analytics metric values corresponding to the second set of raw network analytics metric values; andgenerating a model error measurement based on a comparison between the first calculated set of predicted network analytics metric values and the second set of raw network analytics metric values; selecting, based on a parameter set score, a set of parameters for the selected network analytics forecasting model from a plurality of parameter sets, wherein selecting the set of parameters comprises:
calculating a second set of predicted network analytics metric values using the set of parameters and the selected network analytics forecasting model; andgenerating the parameter set score for the set of parameters based on a difference between the second set of predicted network analytics metric values and the second set of raw network analytics metric values; predicting subsequent network analytics metric values using the selected network analytics forecasting model and the selected set of parameters; determining the effectiveness of the one or more marketing promotions using the predicted subsequent network analytics metric values, the selected network analytics forecasting model, and the selected set of parameters; providing information related to an advertising campaign based on the effectiveness of the one or more marketing promotions.