Data mining techniques for gas price data

点击翻译按钮获取中文摘要

This research project will address the problem of mining knowledge from crowd-sourced, spatio-temporal pricing data, principally concerning retail motor fuel (gas) prices. The industrial partner for this proposal is GB Internet Solutions (GBIS), a growing Regina-based developer of retailer- and consumer-oriented mobile applications (apps) for convenience retailers and consumers. The GBIS consumer app for the gas market is called GasBuddy, and the corresponding website is GasBuddy.com. The research team, which consists of U. of Regina and GBIS personnel, will develop novel techniques for collecting, storing, mining, and visualizing crowd-sourced, spatio-temporal data in the form of gas price reports. We will apply these techniques to the database accumulated by GBIS to gain more insight into the gas market in North America and to enhance the experience of their members by providing more, and better, information and analytics to those using the GasBuddy app. In our research, we will focus on six main problems, as described below. We will design and evaluate algorithms for building predictive gas price models to improve the predictive accuracy of such models while maintaining acceptable computational efficiency; identifying competitors in setting gas prices as an additional component in creating accurate and reliable predictive gas price models; appropriately and accurately filling in missing data; and characterizing gas price setting strategies at various levels of granularity to discover patterns of behaviour and incorporate them into our prediction models. As well, we will focus on the design and evaluation of new visual formats for obtaining and presenting information about gas prices that can be utilized by members to assist them in making informed purchase decisions, and identifying the best methodology for evaluating predictive models derived from spatio-temporal data when initial data exists, as the best combination of building, training and testing is not clear.