Titan 1999: Data Mining & Finite Population Sampling in Marketing

A dynamic system to estimate the nation and state wide monthly retail sales and market share of the company was researched, designed, developed and implemented. Monthly observations on retail sales of multiple brands by a representative sample of dealers of the company, drawn using a stratified probability proportional to size sampling, was proposed to be used for this purpose. Monthly wholesale purchase data of these dealers was mined and the discovered patterns were exploited for both the notion of ``size'' and optimal stratification criteria. Next a fully automated Decision Support System was developed using Visual BASIC and MS-Access for maintenance and upgradation of the dealer database, stratification of the dealers using Hartigan's $k$-means clustering algorithm, optimal sample size allocation to each stratum, drawing of the sample by accommodating various logistic constraints, and finally providing the required monthly estimates.