Combining Stream Mining and Neural Networks for Short Term Delay Prediction

TL;DR

This paper proposes a hybrid delay prediction method for public transport vehicles that combines stream mining with neural networks, enabling real-time predictions without extensive training data and improving accuracy.

Contribution

It introduces a novel hybrid approach using Hoeffding trees and multilayer perceptrons for short-term delay prediction in public transport systems.

Findings

Hybrid method provides real-time predictions.

Neural networks improve prediction accuracy.

Eliminates need for extensive training data.

Abstract

The systems monitoring the location of public transport vehicles rely on wireless transmission. The location readings from GPS-based devices are received with some latency caused by periodical data transmission and temporal problems preventing data transmission. This negatively affects identification of delayed vehicles. The primary objective of the work is to propose short term hybrid delay prediction method. The method relies on adaptive selection of Hoeffding trees, being stream classification technique and multilayer perceptrons. In this way, the hybrid method proposed in this study provides anytime predictions and eliminates the need to collect extensive training data before any predictions can be made. Moreover, the use of neural networks increases the accuracy of the predictions compared with the use of Hoeffding trees only.