Grant: Engage Grants Program

GB Internet Solutions (GBIS) has accumulated gigabytes of historical information on gas prices and is collecting approximately 500,000 updates a day, with 90% of updates entered via the GasBuddy app on a smart phone. They want to use data mining techniques to find patterns in this data, including contrasting sets, association rules, predictions of when prices will go up or down, and probabilities of reported prices being accurate. In this project, we will co-operate with them to enable them to achieve these goals. The main research objective of the project is to design and validate new data mining techniques and interestingness measures that have significant practical value in the summarization and analysis of historical gas pricing information. The project will focus on contrast sets and interestingness measures. Briefly, a contrast set is a set of association rules, each in the form X --> Y (S%), where the Xs are all the same, the Ys describe distinct groups, and the S percentages tell the proportion of examples in each group for which the association rule is true. An interestingness measure is a function that yields a numeric value, typically in between 0 and 1, that attempts to quantify the amount of interest a user will have in a contrast set discovered by data mining software. The project will also compare the effectiveness of the new data mining techniques to older data mining techniques such as statistical inference and association rule mining. GBIS has identified two potential avenues of exploitation. First, since they have reliable records over several years for gas prices at a high percentage of gas stations in North America, they believe that they can market the results obtained by data mining to petroleum companies, automobile manufacturers, etc. Secondly, they believe that they can use data mining to improve the information that they provide to consumers about gas prices.