I not too long ago confronted a problem to assist a Buyer Assist division predict buyer name quantity: what number of telephone calls would possibly arrive subsequent month? The issue: all months usually are not created equal.
Well being insurers face a spike in name quantity throughout open enrollment in November, schooling service suppliers are additional busy when youngsters return to highschool in August, whereas the Journey business’s heaviest durations are the months main into the summer season.
Taking the common of the prior 12 months simply doesn’t make sense for companies like these, however that was precisely how the projections have been at the moment calculated. As well as, the enterprise caters to many alternative industries, every having their very own patterns of seasonal utilization, so I couldn’t merely apply a time collection decomposition throughout the whole dataset.
How can we use machine studying to strategy such a problem?
As a fast evaluation, constructing predictive fashions for time collection information begins by gathering historic information, then…