Replication Data for: Spatial Platoon Context vs. Architectural Complexity: Predicting Headway Degradation in High-Frequency Bus Networks

This repository contains the dataset, evaluation code, and result outputs associated with the study "Spatial Platoon Context vs. Architectural Complexity: Predicting Headway Degradation in High-Frequency Bus Networks."

The core dataset encompasses over five months (October 2025 to March 2026) of continuous, high-resolution General Transit Feed Specification Realtime (GTFS-RT) telemetry from 30 diverse high-frequency bus routes in Melbourne, Australia.

Unlike traditional transit datasets that treat vehicles in isolation, this data has been heavily engineered to include a Platoon Spatial Matrix. Every row captures synchronous 5-minute snapshots of the physical kinematic state of an interacting transit platoon. The target variable is defined as Kinematic Convergence (Δ Gap): the physical change in the spatial gap between consecutive buses exactly 15 minutes into the future.

The repository provides the exact Python framework used to evaluate whether predicting this headway degradation requires deep historical sequence memory (Long Short-Term Memory, Multi-Headed Transformers) or if a simple instantaneous snapshot of the platoon (Deep Neural Networks, Linear Regression) is sufficient.

Keywords (Tags)
Bus Bunching; GTFS-Realtime; Headway Prediction; Deep Learning; Spatial Interaction; Public Transport; Transit Telemetry