Taxi Demand Forecasting: A HEDGE-Based Tessellation Strategy for Improved Accuracy

Abstract

A key problem in location-based modeling and forecasting lies in identifying suitable spatial and temporal resolutions. In particular, judicious spatial partitioning can play a significant role in enhancing the performance of location-based forecasting models. In this work, we investigate two widely used tessellation strategies for partitioning city space, in the context of real-time taxi demand forecasting. Our study compares (i) Geohash tessellation, and (ii) Voronoi tessellation, using two distinct taxi demand datasets, over multiple time scales. For the purpose of comparison, we employ classical time-series tools to model the spatio-temporal demand. Our study finds that the performance of each tessellation strategy is highly dependent on the city geography, spatial distribution of the data, and the time of the day, and that neither strategy is found to perform optimally across the forecast horizon. We propose a hybrid tessellation algorithm that picks the best tessellation strategy at each instant, based on their performance in the recent past. Our hybrid algorithm is a non-stationary variant of the wellknown HEDGE algorithm for choosing the best advice from multiple experts. We show that the hybrid tessellation strategy performs consistently better than either of the two strategies across the data sets considered, at multiple time scales, and with different performance metrics. We achieve an average accuracy of above 80% per km 2 for both datasets considered at 60 minute aggregation levels.

Taxi High-Income Region Recommendation and Spatial Correlation Analysis

“…4. Use the Voronoi subdivision to optimize urban space and use Tyson polygons for higher performance in spatial modeling [41][42][43]. Table 4 describes the number of Tyson polygons in each administrative area.…”

“…Table 7 counts the numbers of the high-income regions and the ID of high-income regions at different periods, so that we can get the high-income regions at different periods. It can be seen from Table 6, there are 10,28,43,13,35,44,12,64, and 71 high-income order hot-spot units at 5:00, 8:00, and 18:00 on October 5, October 9, October13. From Table 7, there are 29,66,74,30,90,74,29,107 and 114 high-income order hot-spot units at 5:00, 8:00, and 18:00 on October 5, October 9, October 13.…”

Short-term Traffic Demand Prediction using Graph Convolutional Neural Networks

“…Instead of fixed ensemble weights, the weights were updated based on the prediction performances of previous time-steps. Davis et al [2] shortlisted a couple of time-series techniques to fit the taxi data. In addition, they developed a multilevel clustering technique that can explore the correlation between adjacent subareas.…”

“…Therefore, forecasting short-term demand throughout the city is of great importance to the systems. Over the past few years, various data analysis models have been proposed to solve the short-term forecasting problem, including time-series analysis models [2,11], machine learning models [6,12], tree-based models [17]. Deep Learning (DL), a particular type of neural network, is a promising methodology, attracting much attention in the transport domain [19,20,23].…”

LSTM-Based Anomaly Detection: Detection Rules from Extreme Value Theory

“…The demand for taxis changes dynamically with daily human mobility patterns, along with other non-periodic events. While short-term taxi demand forecasting models may learn periodic patterns in demand [7], [12], they are normally unable to accurately capture non-periodic mobility events. It is necessary to detect these unusual events as they often indicate useful, and critical information that can yield instructive insights, and help to develop more accurate prediction models and strategies.…”