Regression to identify significant drivers of sales, forming hypothesis with significant factors, then using naive bayesian classifier to test hypothesis is quick and reasonably good approach. Especially so for hub and spoke operations in Singapore where forecasting regional needs are challenging. It can also be used to compare performance between stores and where new stores should be opened.
Naive Bayesian classifier has been found to perform well against decision trees and neural network classifiers. A good example of naive Bayesian classifier, including Laplacian Correction, can be found at http://cis.poly.edu/~mleung/FRE7851/f07/naiveBayesianClassifier.pdf
Marketing budgets may be allocated base on rate of change of probabilities with time, so the above approach helps companies to obtain quantitative comparisons weekly or every 2 weeks. Even for probability in the first pass, the difference in null and alternate hypothesis already helps to clarify the strength of the null or alternate hypothesis.
Edmund,
ReplyDeleteGood article. We focus on providing cost analytics solutions to manufacturing and transportation companies. I would like to invite you to check out this article
http://www.simafore.com/blog/bid/104292/transportation-cost-forecasting-with-naive-aggregate-cost-modeling
which details some of the items you mention.