Time series demand forecasting on NYC taxi dataset

• Main Author: Teh, Timothy Rui Sheng
• Other Authors: Arvind Easwaran
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Time series forecasting
• Online Access: https://hdl.handle.net/10356/180988
• Institution: Nanyang Technological University

This project investigates time series forecasting on the NYC Taxi & Limousine Commission (TLC) dataset, aiming to determine how various data preprocessing techniques and feature engineering methods can improve the accuracy of predictions. The TLC dataset, rich in information on taxi demand and supply, serves as a practical case study for evaluating the effectiveness of these advanced forecasting techniques. Key experiments were conducted using TSfresh for time series feature engineering, enabling the extraction of significant features from the data. Additionally, clustering methods were employed for data aggregation, while Optuna was utilized for hyperparameter tuning to optimize model performance. The best-performing model achieved a Weighted Mean Absolute Percentage Error (WMAPE) of 12.44 by incorporating covariate information such as weather conditions and average fare prices. Although clustering and aggregation methods were explored, they did not yield substantial improvements in accuracy. The project underscores the critical role of data preprocessing and feature engineering in time series forecasting, providing valuable insights for enhancing predictive accuracy in dynamic environments like urban transport. Overall, this research contributes to a better understanding of the methodologies that can lead to more accurate and reliable time series predictions.