Modern telecommunication technology enables service systems to operate on a larger scale than ever. With a large user base, new challenges also emerge in the decision making of these systems. This thesis present two topics that uses data-driver approaches to help the with the decision making in large scale systems.

The first topic focuses on a fraud detection task in the express delivery industry. In the case, the preemptive approach shows superior advantage over the reactive approach. We also found that, by using different under-sampling ratio, the machine learning model can achieve better performance. By considering the economic model and the data pattern, a framework is proposed to detect risky orders in large-scale business.

In the second topic, we study a ride-sharing mode...[ Read more ]

Modern telecommunication technology enables service systems to operate on a larger scale than ever. With a large user base, new challenges also emerge in the decision making of these systems. This thesis present two topics that uses data-driver approaches to help the with the decision making in large scale systems.

The first topic focuses on a fraud detection task in the express delivery industry. In the case, the preemptive approach shows superior advantage over the reactive approach. We also found that, by using different under-sampling ratio, the machine learning model can achieve better performance. By considering the economic model and the data pattern, a framework is proposed to detect risky orders in large-scale business.

In the second topic, we study a ride-sharing model. With numerical approach, we identify a key relationship between the expected pick-up distance and the number of drivers and passengers waiting to be matched. With extensive simulation and a two-phase queueing model, we investigate on how ride-sharing platforms can tune the matching radius to make order-dispatching decision efficiently.