Problem Statement

Big Night Restaurants & Bars (BN) is looking for the opportunity to offer on-line ordering. The idea is to be able to offer intelligent suggestions to the customer comparable to those given by good wait staff. To facilitate this, the online system will offer meal suggestions considering variables such as: the cost of ingredients, inventory availability, the probability of customer acceptance and the overall appeal of the combination. The system must be able to handle the ordering process from a regular menu as well as from a catering menu.

Solution Strategy

Team Light Blue proposed that an online ordering system could be created with a user interface which would include some additional key features. Users would be able to sign in via any well-known social networking service. Food suggestions would be offered to the customer based upon past order and browser history, social network data, and expert knowledge gleaned from the chefs, servers, and managers of the subsidiary restaurants. These suggestions would be offered in a subtle and innovative manner, designed to maximize their effectiveness. The interface would be easy to use and would be mobile-friendly. Estimated time for pickup could be computed based upon the items ordered and how busy the restaurant is and would be provided to the customer on-screen after they place their order.

Problem Statement

Autodata implements a wide range of software used by millions for marketing the automotive industry, mostly within North America. The huge amount of click stream data generated each day is a valuable source of marketing information. The problem is determining how to efficiently collect and analyze these immense sets of data and combine them into one, easy to mine format. Sorting the information in this way would help to assess customer needs and aid them in their selections.

Solution Strategy

Team Sage suggested that the best approach would begin with the collection of the data itself by cleaning the sets of data without any of the unnecessary and unwanted information such as incomplete sessions, replication records, etc. and then standardizing into one uniform format.

Team Sage proposed two different methods for addressing different questions. By mapping the sequences to graphs, Autodata could find out information such as; the most favorable companies and vehicles, the most favorite options for each vehicle, parts of the software with a possible conflict and the most important competitor for each company. Second, by mapping the sequences to regular expressions other questions such as 'what can cause the customer to change the company?', 'what percentage of the customers have a successful experience working with the software?' or 'what are the possible reasons making the customers to log out or exit the system before achieving their goals?' can be answered.

Problem Statement

Environics Analytics uses various data for its marketing and analytical services. However, sometimes necessary data is unavailable. Certain dissemination areas are declared to have missing average household income by Statistics Canada. We need to estimate the average household incomes for these areas from the provided 2006 census demographic data.

Solution Strategy

Team Yellow used two phases to solve this problem. They first estimated missing demographic information by using an approach known as k-Nearest Neighbour. They estimated missing average household income based on a weighted average of neighbouring regions' average household incomes.

The second phase used a multifaceted approach. First, Team Yellow used general linear regression (i.e. line of best fit) to determine which demographic factors are relevant and create a mathematical model estimating income based on the most relevant factors. Second, they once again used the k-Nearest Neighbour approach, assuming that a given region has similar demographic factors to its neighbouring regions. The final approach used the model from the first phase to estimate the missing income using the estimated demographic factors in the second phase. Testing the model showed encouraging results.

Problem Statement

No precise definition of urbanity exists, however it intuitively covers domains such as commercial development, population density and proximity to a city centre. The task was to develop a methodology for estimating the urbanity of a given Dissemination Area (DA) of the Canadian census. Accurate 'urbanity scores' have wide applicability in the domains of marketing, city planning, urban development.

Determining these urbanity scores is non-trivial, and traditionally they are highly subjective; a methodology that can estimate the urbanity score of any DA objectively will result in a better planned, better marketed, smarter city.

Solution Strategy

Team White proposed a strategy using the principles of machine learning to provide an objective measure of urbanity. A continuous restricted Boltzmann machine was trained on a small dataset containing relevant demographic and geophysical statistics for both highly urban and highly rural examples. With the training complete, the Boltzmann machine was able estimate the urbanity score of a novel DA (not included in the training set) based on its demographic and geophysical statistics. As such, Team White provided a continuous measure of urbanity reflecting only the statistical trends present in the entire dataset.

Problem Statement

The Ontario Municipal Benchmarking Iniitative (OMBI) collects over 900 measures from 16 different municipalities. However, this increasingly large amount of data is difficult to synthesize and understand. We require an innovative yet feasible strategy for OMBI to present these data to better inform decision-makers.

Solution Strategy

Team Orange recommended that the raw data from excel should be manipulated into a working SQL-like database. The next step would be to apply modern statistics to create indexes of different measures and compare cities. Finally, Team Orange suggested OMBI apply an innovative design to improve the visualization of the data through either Infographics or Interactive Maps.

Problem Statement

Two databases with imperfect index function must be joined in order to plot water usage data against geographic information. In particular, the imperfection arises from human introduced error of addresses formulated as:

Unit Number StreetName StreetType StreetDirection for example A 123 Fake Ave N.

Manual correction of the addresses is possible but time consuming. The central task is to automate the correction of these errors.

Solution Strategy

The Green Team's foremost recommendation was to eliminate the human error by standardizing the entry of the addresses (especially the Units which are by far the most prevalent type of error).

To clean the current database the Green Team recommended the application of standard spell checking technology to do a one-time correction. The Statistical Analysis Software (SAS) provides a library function SoundsLike which is a robust, industrial quality solution to this problem. Alternatively, a software programmer of sufficient skill could implement a technique similar to that used in bioinformatics called sequence alignment which is a way of arranging sequences to identify regions of similarity, but this would be slower and potentially less accurate (as it depends on the quality of the implementation). Finally, as the correction is for a small database (of hundreds of thousands of entries), efficiency considerations should be ignored.

ipsw - at - uwo.ca