Data Acquisition for Business Analytics

We study data acquisition for business analytics considering both data quality and acquisition cost. We propose using a divide-and-conquer technique to find empirical distribution of the data and then using auctions to solicit data based on the distribution. To determine how many customer records to acquire, we formulate the acquisition problem as an optimization problem that maximizes the quality of the acquired data while keeping the acquisition cost as low as possible. For descriptive analytics, we derive the closed form solution for the optimization problem, which finds the amount of data to acquire that best represents the distribution of the target population relative to the acquisition cost. For regression-based predictive analytics, we formulate the acquisition problem as a mathematical programming problem that can be solved efficiently; the optimal solution minimizes prediction error and acquisition cost. An experimental study has been conducted to demonstrate the effectiveness of our approach