Grant: Discovery Grants Program - Individual

Modern data often exhibit complex characteristics that necessitate the development of sophisticated methods. Firstly, the relationship between covariates and the response variable is often intricate, involving non-linear effects and interactions. This implies that flexible models capable of automatically uncovering these relationships are useful. Secondly, the data may exhibit dependency structures (e.g., network or clusters) and dynamic (temporal) patterns, which pose additional challenges for analysis. Hence, it becomes crucial to extend these flexible methods to effectively analyze such complex data. One such method, random forest, is very popular among data analysts and widely used in practice. This popularity stems from its exceptional performance-to-complexity ratio. Furthermore, random forests are inherently parallelizable, making them practical for handling large data sets. The method's renown is not only attributed to its impressive empirical performance across diverse real-world scenarios but is also substantiated by recent theoretical advancements. With their numerous advantages, random forests have found adaptation in some settings dealing with complex data. Nonetheless, several challenges persist, and certain problems have yet to be addressed using these methods, as elaborated in the detailed research agenda below. Consequently, the long-term objective of this research program is to develop novel methods based on random forests for the analysis of data featuring complex dynamics and dependence structures. The research program is divided into two axes: 1)Prediction and imputation of traffic volume using random forests 2)Random forests for survival, longitudinal and clustered data. The research questions in the first axis are motivated by applications in transportation, but the methods will be general and applicable in other fields as well. The methods developed in the second axis have applications in several fields (e.g., engineering, medicine and social sciences) and are partly motivated by issues encountered in various industries through the supervision of graduate students.