Data Scientist, working on optimization algorithms and data analysis at Milanamos, a specialist in data science for travel industry.
PhD in Computer Science and Industrial Engineer graduated from Ecole Centrale Paris, major in Operational Research applied on logistics and transportation systems. I also hold a Master degree in Big data/Business analytics delivered by The University Paris-Dauphine. Interested in Operational Research, Artificial Intelligence, and Data Science.


Air transportation network optimization: market analysis and selection.


In the airline industry, airline network planning problems are various and complicated. Solving these problems aim at reducing costs and maximizing revenues. The revenues can be increased by improving the quality of service. Consequently, the company can be more competitive. One way to improve the competitiveness is to catch new passengers on existing flight connections or on new markets.
The new markets selection is the key to company decision-making process. It consists in determining network structure to operate, potential markets selection to deserve, passengers flow, their choice of itineraries as well as incomes and costs involved in these decisions.
The work carried out in this industrial thesis for Milanamos, a company specializing in information technologies applied to travel industry, is aimed at improving its market planner engine. Milanamos is developing an application for the analysis and simulation of markets intended for airports and airlines. It offers its customers a decision-making tool to find an economic opportunity by analyzing data history, scheduling flights, and visualizing the network etc.. This project takes place earlier in the decision process. It consists of two phases: improving the visualization and calculating the market shares.
We started by analyzing data from Milanamos and modeling the air transport network with a time-independent graph and storing it in Neo4j graph database. The The Flight Radius problem is a multicriteria problem. It consists in determining a subgraph of realistic paths which limit, for example, loss time or additional cost compared to the best paths in the subgraph. Several methods have been proposed based on queries and on shortest path algorithms coupled with acceleration and parallelism techniques. The results demonstrate that the algorithm based on Dijkstra is the most efficient.
The second step consists in calculating market shares for the subgraph. In practice, the calculating process enumerates all itineraries of the considered market. Furthermore, other markets are concerned by the new service. A method has been proposed to reduce a combinatorial on itineraries to evaluate and to consider other markets. The new approach is based on Hub Labeling algorithm and Logit model to estimate itinerary market share.