07/10/2023
By Sohil Khokhar

Richard A. Miner School of Computer & Information Sciences, Department of Computer Science, invites you to attend a Master's Thesis defense by Sohil Khokhar on "Learn Graph Machine Learning To Predict Trip Time."

Candidate Name: Sohil Khokhar
Degree: Master's
Defense Date: Friday, July 21, 2023
Time: 11 a.m. to noon
Location: Via Zoom 

Advisor: Tingjian Ge, Department Of Computer Science, University of Massachusetts Lowell

Committee Members:

  • Cindy Chen, Department Of Computer Science, University of Massachusetts Lowell
  • Yimin (Ian) Chen, Department Of Computer Science, University of Massachusetts Lowell

Brief Abstract:

Accurately predicting trip durations in large-scale transportation datasets is crucial with wide-ranging applications, including optimizing routes, forecasting demand, and managing real-time traffic. In this thesis, we explore the application of graph machine learning techniques to enhance trip duration prediction in a vast New York taxi dataset. By leveraging the inherent spatial and temporal relationships among taxi trips, we construct a graph representation of the dataset and employ advanced machine learning algorithms to model and predict trip durations.

The initial phase of our research involves preprocessing the dataset and constructing a graph where each trip is represented as a node while the edges capture the relationships between trips. Extracting pertinent features from the dataset, such as pickup and drop-off locations, timestamps, and other trip attributes, enables us to create informative feature vectors for each node in the graph. Next, we employ graph machine learning algorithms to capture the complex dependencies and patterns within the dataset effectively. By considering the neighboring trips and their associated features, our models leverage the graph structure to learn contextual information and make more accurate predictions. We employ graph convolutional networks (GCNs), Graph Attention Networks (GATs), and GraphSage for node-level and graph-level predictions, leveraging the wealth of information encoded in the graph.

To evaluate the effectiveness of our approach, we conduct extensive experiments on the large New York taxi dataset. We compare the performance of our graph machine learning models with traditional machine learning methods, assessing their predictive accuracy, efficiency, and scalability. Additionally, we analyze the impact of different graph construction strategies and feature representations on prediction performance. The results demonstrate the significant enhancement in trip duration prediction accuracy achieved by our proposed graph machine learning approach compared to traditional methods. By harnessing the power of graph-based techniques, we effectively exploit the spatial and temporal relationships, leading to more precise and context-aware predictions. Furthermore, our models exhibit robustness and scalability, enabling their application to large-scale transportation datasets.

In conclusion, this thesis presents a novel approach that leverages graph machine learning for predicting trip durations in a vast New York taxi dataset. The integration of graph structures and advanced learning algorithms offers a powerful framework that improves trip duration predictions, facilitating more efficient transportation planning and decision-making.