Search

The quality of masters’ dissertations is an important index of graduate education, which can be in part reflected through the grades given by experts. This study aims to find the factors positively correlated to the grades, and then use them to predict the grades and quality of dissertations. We applied four typical machine learning models to calculate the impacts of several factors extracted from the contents of dissertations on the grades. It shows that the random forest model outperforms logistic regression, support vector machine, and naive Bayes on recognizing the dissertations with a high grade. It also shows that the quantity of publications is the most important predictor to the grades, compared with the quantity of publications, the length of dissertations, the quantity and quality of references. And the quality of references is a significant predictor of producing high quality publications. Those findings can be utilized to predict and recognize high quality dissertations.

Human reaction time has a substantial effect on modeling of human behavior at a microscopic level. Drivers and pedestrian do not react to an event instantaneously; rather, they take time to perceive the event, process the information, decide on a response and finally enact their decision. All these processes introduce delay. As human movement is simulated at increasingly fine-grained resolutions, it becomes critical to consider the delay due to reaction time if one is to achieve accurate results. Most existing simulators over-simplify the reaction time implementation to reduce computational overhead and memory requirements. In this paper, we detail the framework which we are developing within the SimMobility Short Term Simulator (a microscopic traffic simulator), which is capable of explicitly modeling reaction time for each person in a detailed, flexible manner. This framework will enable modelers to set realistic reaction time values, relying on the simulator to handle implementation and optimization considerations. Following this, we report our findings demonstrating the impact of reaction time on traffic dynamics within several simulation scenarios. The findings indicate that in the incorporation of reaction time within microscopic simulations improves the traffic dynamics that produces more realistic traffic condition., Singapore-MIT Alliance for Research and Technology