Machine Learning for Transportation Research
In this research stream we use machine learning algorithms to analyze various data sets to better understand the travel and activity patterns of individuals and specific segments. We use clustering algorithms, among others from the classical K-means to deep clustering with Variational Autoencoders, as they are more suitable for detecting mobility patterns of group of individuals with similar mobility behaviour such as age, ethnicity, geographic location etc. Such an analysis can reveal unique patterns and key characteristics in the activity sequence of different groups which will allow the customization and personalization of mobility services to the groups of interest. Insights gleaned from this analysis of actual travel and activity behavior can provide guidance on designing various mobility services' packages and menu. On top of that we use machine learning algorithms for calibration as hundreds of parameters are involved in the calibration of Activity-Based models focused on behavioral theory, to properly frame the required detailed socio-economical characteristics. To address these challenges, we use a novel Bayesian Optimization approach that incorporates a surrogate model defined as an improved Random Forest to automate the calibration process of the behavioral parameters.

K-means geospatial clustering

Hierarchical clustering