Sponsor:  Laser Tag Source

Cadet:   Joe Bobay
Advisors:  Ben Grannan and John David

Laser Tag Source is a laser tag rental company in Lynchburg, VA. They ship equipment to all 50 states in the US through UPS by ground. The problem that they presented us with is that they have spent about $60,000 in shipping alone last year and that shipping something to California from Lynchburg and back takes 13 days. They want us to find out if it is profitable to open a second distribution center, and find out exactly where that should be, the end goals being to reduce shipping costs and to increase availability of merchandise.

Sponsor:  Society for Industrial and Applied Mathematics (SIAM)

Cadets:  Austin States, Patrick Lenehan
Advisor:  Jessica Libertini

This primary goal of this project was to help assess and make recommendations for improving the judging of a SIAM-sponsored high-school mathematical modeling competition, the Moody's Mega Math Challenge (also called M3). The M3 challenge uses hundreds of volunteer judges, mostly faculty from across the country, to assist in a triage round of judging, aiming at identifying which papers should advance into the final rounds of judging. Due to the disparity of judging scores, as individual judges bring their own biases, a calibration system is currently in place. Over the years, various calibration algorithms have been applied, but their effectiveness has been unmeasured. In this project we examined the effects of past, current, and proposed calibration systems, as well as calibration systems used by similar competitions. After analyzing these systems, the students made recommendations for improving the systems, as well as recommendations for future research.

Cadet:  Connor Loken

Advisor:  John David

This work looked at predicting all NCAA football and basketball games between 2012 and 2014 using a novel approach. The inputs for the prediction were the Las Vegas line along with the amount of web traffic for each team. We then used an adaptive neuro-fuzzy inferential system to predict the outcome of the game. This approach combines a neural network approach with a fuzzy logic membership function approach which proved to be extremely accurate in predicting the winners of all games.

Sponsor: Johns Hopkins University Applied Physics Lab

Cadet Researchers: Stephen Mascioli, Cameron Armstrong, Ryan Harner

Faculty Advisor: John David

This project was initiated to better understand the capabilities of naval sea and air assets off the Horn of Africa to be used as a tool against piracy in that region. A model was developed to simulate possible plane routes and pirate detection probabilities based off of realistic operating conditions in order to determine the maximum possible area coverage in a particular region over 7 days. Major factors in the model were the density of merchant ships in the region and the availability of patrolling assets.

Sponsor: Valley Program for Aging Services (VPAS)

Cadet Researcher: Alex Falcetti
High School Interns: Stevan Hall-Mejia
Advisors: Nate Axvig and John David

One of the services VPAS provides is a program in which hot meals are delivered to homebound senior citizens for their noon meal. Currently, over 400 meals are served each day.

Our task was to find a more efficient way to route delivery drivers given the daily fluctuations in both volunteer aid as well as meal demand. To accomplish this, we designed an easy-to-use piece of software that employs a combination of genetic algorithms, allowing VPAS to deliver their meals in a more efficient manner.

Sponsor: City of Lexington
Cadet Researcher: Robert “Chap” Michie
High School Interns: Kunal Ghandi and Stevan Hall-Mejia
Faculty Mentor: Geoffrey Cox

The City of Lexington is located in Rockbridge County, Virginia. The city owns and operates 27 buildings with 42 separate energy accounts: 13 natural gas and 29 electric. Prior to our project, all records of these accounts were handwritten documents kept within city hall, penciled in by a city employee.

The goals of the project were to first take this information and put it in a digital state and then provide a tool to analyze the data in question. Our tool allows the city to quickly access visual models of each building’s energy consumption so that improvements can be made and inefficiencies addressed. By looking at the visual models, the city can increase its overall sustainability, providing economic, environmental, and social benefits to the community. 

Sponsor: VMI Summer Undergraduate Research Initiative
Cadet Researcher: Will Lucas
Faculty Mentor: John David

Data Mining has become an increasingly popular technique in science, economics, business and sports, especially a sport with as rich data as baseball. In this project we mined millions of baseball statistics to predict which teams will win on a day to day basis as well as over entire seasons.

We focused on statistical techniques of regression, artificial neural networks and decision trees.

Sponsor: National Science Foundation
Researcher: Robert Michie
Advisor: Atin Basuchoudhary and John David

This project examines factors of political risk and their ability to influence and predict real interest rates. Various data mining techniques were used on the ICRG researcher dataset for political risk, which is made up of 12 weighted components.

 The dataset contains data from the years 1984 to 2008 for 37 countries. Unsupervised learning techniques were used in order to find hidden structures within the data itself, while supervised learning techniques were used to make future predictions of the real interest rate.  

 The key points of this work are that money growth is not a good predictor of real interest rates, as well as the fact that a committee of artificial neural networks proved to best the strongest model in terms of predictive performance, and finally that there are certain political risk factors that are indicative of changes in real interest rates.

In this project we explore the importance of data in the process of making predictions of outcomes of National Basketball Association (NBA) games.

We use NBA statistics for the 2004-2005 season up to and including the 2010-2011 season. We utilize unsupervised learning techniques to gain an informed understanding of our data and how it operates. These techniques include Clustering, Principal Component Analysis, and Factor Analysis. We learn predictive capabilities and the different ways in which the statistics impact the outcome of an NBA game. We also use supervised learning techniques to build models in order to make predictions on the score differential between two teams in a given game.

Committees of Artificial Neural Networks (ANN’s) and Forests of Decision Trees are built based on data leading up to seasons we wish to predict. In this study the most accurate models are built using data from all seasons, 2004-2009, and then evaluated using 2010 data to make predictions. Our most accurate model predicted 2010-2011 NBA game winners with an average accuracy of 72.45% and a mean error of 8.32 points.

Cadet Researchers: William Lucas and Steven Geyer
Advisor: Nate Axvig

Different data mining techniques were used to assess the data that we received from the client. The two main techniques used to analyze the data were artificial neural networks and multiple regression analysis.

The goal of the project was to be able to predict errors on a given order. The two techniques yielded comparable results, but due to the complexity of the neural networks we decided to use the multiple regression model to present to our client because it is much simpler to for them to use.

The United Way of Greater Augusta strove to build a database containing facts and figures specific to greater Augusta County, as well as the Virginia comparisons, with emphasis on making it easily updatable for United Way employees and quickly navigable for any member of the community interested in browsing local data.

Another goal of this project is to pinpoint not only the main problems in the community, but connecting the factors that are causing them. In that way, support will be used to resolve the cause of the problem, not a symptom of it.

The database, which included data analysis such as multiple linear regressions and statistical hypothesis tests, could identify what the symptoms are and begin to link some of the factors in order to discover the causes.

Sponsor: Vector Industries
Cadet Researchers: Alexander Lin
Advisor: Geoffrey Cox

Vector Industries is a non-profit employment oriented organization that works to provide individuals with disabilities financial and social independence. Their services include providing complex product assembly, packaging, product testing, and much more.
Recently, Vector has decided to make efforts in transforming specific business operations and processes to ultimately improve their organization's performance. These processes involve the analysis of large amounts of financial data, employee wages, and transportation routes for picking up employees.

To assist in their efforts, a MATLAB transportation routing program was utilized for reducing transportation expenses. Additionally, Microsoft Excel automation tools were developed that conserves time, decreases human error, and generates organized data sets.

Sponsor: VMI and Young Life
Advisor: John David

This project was initiated to better the food buying practices of the Young Life Organization as well as to develop tools for them to utilize in this process. One tool analyzed past food invoices to determine when certain products will historically be the cheapest.

In addition tools were developed to help Young Life build a larger database allowing for their model to become better overtime and to normalize the price lists.

Cadet: Jack Zippel
Faculty Advisor: Maj. John David
Sponsor: Davidson & Garrard, Inc., David Hansen

The goal of this project was to value a set of exchange traded funds (ETFs) that Davidson & Garrard may use as indicators of market potential. Predicting the prices of the ETFs allows for an effective valuation technique: if the price will increase, the ETF is currently undervalued and vice versa.

Predictions spanning two years provide optimal results given that short term market oscillations distort greater trends and larger economic events drastically change long term market direction.

Cadet: Ryan Poffenbarger
Faculty Advisor: Maj. Geoff Cox
Sponsor: American Shakespeare Center

This projects analyzed 13 years of patron and donor data with the primary interest being to understand seating preferences, customer retention and demographics, and donation patterns.

Through summarizing hundreds of thousands of ticket sales in simple statistics and graphical representation of these we were to give the Center clear quantitative evidence for how to better market and grow donations and ticket sales.

Cadets: Austin States and Josh Grant
Faculty Advisor: Maj. Randy Cone

This project worked for a local manufacturing firm to help them develop a quantitative tool to better understand their production process and capability.

This tool gives guidance on how to flex production to produce orders in an efficient and timely manner while considering variation in product demand, multiple machines in the production process with varying capacity and reliability and variation in product production time.