University of Twente Student Theses
Identification of business travelers through clustering algorithms
Piggott, J.J.H. (2015) Identification of business travelers through clustering algorithms.
This is the latest version of this item.
|
PDF
2MB |
| Abstract: | Over the fiscal year 2014 Air France-KLM reported a net loss of €198 million (CAPA, 2015). Long-haul flights are a market in which the combined group has performed well in the past, yet it faces increasing competition (Skift, 2015). Efforts to compete on short-haul flights with established Low Cost Airlines (LCA) have proceeded slowly due to labour related problems with Transavia, the groups own LCA. To better compete with LCA's, Middle East and Far East airlines as well as improve their operational profits Air France-KLM needs to better understand its passengers and their desires. Traditionally the business travel segment has been the group’s most profitable segment. Previous market research has shown that only half such travelers have a corporate contract with Air France-KLM. This suggests that if Air France-KLM is able to identify which passenger is a business traveler it could foreseeably improve its operational results. If KLM can better understand the needs of their passengers they will be better able to target them with sharper pricing of tickets, increase client retention with frequent flyer programs and improve ancillary revenue. The goal of this research effort is to discover a better market segmentation model suitable for use by Air France-KLM. In order to achieve the goals of better understanding airline passengers this research effort uses actual passenger flight movement data. Records were collated and grouped such that for each identifiable unique passenger a record exists of all their flight movements within a year. Metrics such as frequency of travel, distance traveled and the weight of baggage checked in are 3 among more than 25 variables that permit passenger behavior to be identified and passengers grouped together. This research effort used machine learning techniques aimed at recognizing airline passenger behavior and grouping them together. The theory behind techniques such as supervised and unsupervised learning is discussed. Practical issues with unsupervised learning algorithms such as K-means, Expectation-Maximization and Hierarchical Clustering Algorithms are also detailed. Such algorithms make it possible to group data points that share similar features together. Such clusters are then identified according to existing airline market segments and interviews with stakeholders. For supervised learning algorithms such as Decision Tree (CART) is explained. Supervised learning makes it possible to create a predictive model of labeled data. It is thus possible to describe when an airline passenger, as an example, should be recognized as a business traveler. This thesis expands this effort by also using semi-supervised learning. This technique has the advantage in that it requires only a sample of the data to be labeled for use in training an algorithm. Semi-supervised learning offers the possibility of being more accurate than unsupervised learning without having to incur the cost associated with labeling the data set. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Business Information Technology MSc (60025) |
| Link to this item: | https://purl.utwente.nl/essays/67067 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page