The chief marketing officer (CMO) at an automobile agency was looking at a list of car model features, which he had received from the manufacturing plant. He was expected to provide the manufacturer’s suggested retail prices of the cars to dealers the following week and had to decide on the base prices. The CMO asked a data scientist at the research lab to predict prices using the data of past car models. Each car model had different features that could affect the price. The data scientist decided to use feed-forward neural networks as a tool for predicting the prices of new models. After comparing different prediction models, he also wanted to determine which prediction model was suitable for car manufacturing plants.
This note explains the fundamentals of neural networks that are used for classification and prediction systems, with a primary focus on feed-forward back-propagation neural networks.
A management science professor had an unpleasant experience with a hotel she stayed at in New York City. Consequently, she wanted to figure out if hotel ratings were enough to recommend a hotel, or if customers’ text reviews could be used as more important and accurate indicators of customers’ hotel experiences. The exercise serves as an introduction to the topic of text analytics—specifically, sentiment analysis—and introduces the concept of text mining and the importance of dealing with unstructured datasets. Much of the exercise focuses on the method and rationale behind document indexing and the subsequent weighting of the indexed terms through term frequency–inverse document frequency. Textual data from customers’ hotel reviews are provided to apply the text mining techniques and to provide insight for a better decision-making process that would help the professor in her next hotel booking.
Queues (or waiting lines) are common in modern life. We wait in lines at the campus cafeteria, to board an airplane, in the emergency room, and when we call a company for customer service. We even wait for applications to be initiated and processed by our smartphone processors, and our online orders wait to be fulfilled at companies’ warehouses before being shipped to us. How long we wait in line depends on several factors and parameters. It may depend on the number of servers working and the amount of time it takes to serve each individual customer (the system’s capacity) or on how many customers arrive at any given time (the system’s demand or flow rate). Most importantly, queues form due to random variations in arrival patterns and service times.
Hospitals frequently deal with congestion and blockage that affects patient flow and increased costs. In 2018, at University Hospital, part of South West Health Centre, patient flow in the medical surgical intensive care unit seemed to be highly susceptible to congestion, and this was creating ripple effects throughout the hospital and leading to increased costs. The complexity of patient flow presented an opportunity to build a discrete-event simulation model to provide insights regarding the patient flow, costs, and opportunities for improvement. But how could the hospital’s management use a simulation model of patient flow to determine alternatives to optimize both patient flow and costs?
In 2017, while donating blood on campus, a management science graduate student noticed that the mobile blood donor clinic set up at his university’s community centre was a congested tandem queuing system. Finding one-and-one-half hours too long for donors to wait, the student considered how the process could be reduced by at least half an hour. He needed to devise a reasonably precise model to represent the donor flow in the clinic. Using either the mode or the average service times supplied by the nurses, the student could build a relatively straightforward discrete-event simulation model to identify bottlenecks and improve the donor flow.