The thesis develops an automated system for monitoring bus trips based on AVM (Automatic Vehicle Monitoring) data, with the goal of supporting the Operations department of Autoguidovie while remaining scalable across all companies in the group, such as Cavourese, in the timely detection of service anomalies. The specific issue addressed concerns unregistered trips, which account for about 10% of the total and represent the main weakness of the AVM information flow. Through an in depth analysis of operational and planning data, integrated with additional corporate sources, the work constructs a set of interpretable features that describe the historical behavior of vehicles, drivers, and routes according to a strictly, past logic. The thesis proposes two complementary tools: (1) an automatic anomaly, detection engine designed to compare planned versus actual operations and generate notifications in case of significant deviations; (2) a supervised predictive model based on logistic regression and trained through down sampling, capable of identifying potentially anomalous trips before they occur. The model, built as an ensemble of five logistic regressions, achieves moderate performance, confirming its ability to capture recurring patterns despite the absence of true causal labels. The data analysis shows that certain anomalies tend to be associated with specific vehicles and drivers, suggesting concrete opportunities for targeted operational intervention. Future integration with additional data sources, such as fault reports, and real time feeds could further enhance the system’s effectiveness both in monitoring and prediction. Overall, the work demonstrates that a data-driven approach, based on interpretable ML and a structured analytical process, can significantly contribute to improving service quality, reducing disruptions, and modernizing operational activities in local public transport
La tesi sviluppa un sistema automatizzato per il monitoraggio delle corse degli autobus basato sui dati AVM (Automatic Vehicle Monitoring), con l’obiettivo di supportare le Operations di Autoguidovie, ma scalabile su tutte le realtà del gruppo come Cavourese nell’individuazione tempestiva delle anomalie di servizio. Il problema affrontato riguarda in particolare le corse non registrate, che rappresentano circa il 10% del totale e costituiscono la principale criticità del flusso informativo AVM. Attraverso un’analisi approfondita dei dati operativi e di pianificazione, integrati con ulteriori fonti aziendali, viene costruito un set di feature interpretabili che descrivono il comportamento storico di mezzi, conducenti e linee secondo una logica strictly‑past. La tesi propone due strumenti complementari: (1) un motore di rilevamento automatico delle anomalie, progettato per confrontare pianificato e realizzato e generare notifiche in caso di scostamenti significativi; (2) un modello predittivo supervisionato, basato su regressione logistica e addestrato tramite downsampling, in grado di individuare le corse potenzialmente anomale prima che si verifichino. Il modello, composto da un ensemble di cinque regressioni logistiche prestazioni moderate confermando la capacità di cogliere pattern ricorrenti nonostante l’assenza di vere etichette causali. L’analisi dei dati evidenzia che c’è una tendenza di alcune anomalie ad essere legate a specifici mezzi e conducenti, suggerendo margini concreti di intervento operativo mirato. L’integrazione futura con dati aggiuntivi (come le segnalazioni di guasto) e con fonti real‑time potrebbe accrescere l’efficacia del sistema sia nel monitoraggio sia nella previsione. Nel complesso, il lavoro dimostra che un approccio data‑driven, basato su ML interpretabile e su un processo strutturato di analisi, può contribuire in modo significativo al miglioramento della qualità del servizio, alla riduzione dei disservizi e alla modernizzazione delle attività operative nel trasporto pubblico locale.
Development of an automated verification system for public transport trips using AVM data
ZANOTTO, GIOVANNI MARIA
2024/2025
Abstract
The thesis develops an automated system for monitoring bus trips based on AVM (Automatic Vehicle Monitoring) data, with the goal of supporting the Operations department of Autoguidovie while remaining scalable across all companies in the group, such as Cavourese, in the timely detection of service anomalies. The specific issue addressed concerns unregistered trips, which account for about 10% of the total and represent the main weakness of the AVM information flow. Through an in depth analysis of operational and planning data, integrated with additional corporate sources, the work constructs a set of interpretable features that describe the historical behavior of vehicles, drivers, and routes according to a strictly, past logic. The thesis proposes two complementary tools: (1) an automatic anomaly, detection engine designed to compare planned versus actual operations and generate notifications in case of significant deviations; (2) a supervised predictive model based on logistic regression and trained through down sampling, capable of identifying potentially anomalous trips before they occur. The model, built as an ensemble of five logistic regressions, achieves moderate performance, confirming its ability to capture recurring patterns despite the absence of true causal labels. The data analysis shows that certain anomalies tend to be associated with specific vehicles and drivers, suggesting concrete opportunities for targeted operational intervention. Future integration with additional data sources, such as fault reports, and real time feeds could further enhance the system’s effectiveness both in monitoring and prediction. Overall, the work demonstrates that a data-driven approach, based on interpretable ML and a structured analytical process, can significantly contribute to improving service quality, reducing disruptions, and modernizing operational activities in local public transport| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_02_Zanotto_Thesis.pdf
accessibile in internet per tutti
Dimensione
2.05 MB
Formato
Adobe PDF
|
2.05 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/253298