Data modelling

As a part of course work, I had opportunity to take following courses which reinforced by concepts in this domain:

  • Transportation engineering
    • Discrete Choice Analysis
    • Urban Transportation Models (mainly covered optimization)
  • Computer science
    • Math for Intelligent Systems (math for machine learning)
    • Machine Learning

Selected research work

My PhD work focuses on extracting travel behavior from Artificial Neural Networks. I have listed some of the research work:

1. Challenges in extracting value of travel time from Artificial Neural Networks

As a part of PhD, I am trying to extract travel behavior from ANNs/deeplearning models and compare it with traditional discrete choice models. In this effort, I extracted value of travel times distribution and found them to be very sensitive to hyperparameters, which is expected and has been reported in the literature. I developed curiosity to understand the cause for non-uniqueness in the value of travel time.

We trained over 243,000 ANN models 243 ANNs with different hyperparameters, each trained 100 times for the same datasets to account for the local minima and random weight initialization. Hi-performance computing was used to parallelize the training process over 1000 CPUs. Later, we diagnosed models and value of travel times using. The output (not discussed in the post) gave us a fair understanding of the causes for

The work is nearly finished and I am preparing the manuscript to be submitted in Transportation Research Part B. I will be sharing the pre-print manuscript here very soon.

2. Analysis and modeling of changes in online shopping behavior due to COVID-19

I utilized National Household Travel Survey (NHTS) data and applied ordered response models to understand the online shopping behavior in Florida. Later, we incorporated these statistically significant attributes in our online survey questionnaire and discrete choice models to understand the impact of COVID-19 in online and in-store shopping behavior.

My role was performing exploratory data analysis (EDA) and modelling both datasets – NHTS and online survey data.

This work has been accepted for publication in Transportation Policy journal.

3. Generating synthetic mode choice data using discrete choice models

As a part of my PhD research, I have generated several synthetic mode choice datasets using discrete choice models. These synthetic datasets have known characteristics which were used to test deeplearning (machine learning) models, in a series of computational experiments, if they can uncover the data generation process.