Cracking the Code on Short Ticketing: Using AI to Tackle a Hidden Fare Evasion Challenge

Have you ever wondered how some transport riders manage to beat the system without jumping a gate? 

As part of a collaboration between Cubic Transportation Systems and independent researchers from Imperial College London, we set out to investigate a subtle, yet costly practice called short ticketing.

Unlike passengers who skip payment altogether by jumping barriers, slithering under the gate, or even pushing through the paddles; short ticketing involves a subtler practice of purchasing a cheaper, shorter distance ticket, but secretly travelling farther than allowed. 

For example, a rider on a bus taps off earlier than their intended destination but stays on the bus anyway effectively avoiding their full fare. Think of it as booking half a trip but riding the full route.

Short ticketing is incredibly problematic for transit agencies. In the UK alone, fare evasion costs the UK rail system an estimated £240 million every year, with short ticketing representing an often undetected aspect of the broader issue. Not only does it cost agencies revenue, it also skews ridership data, making it harder for agencies to plan services and set fair fares.

How Short Ticketing Works

Short ticketing at its core is the intentional under-purchasing of your transit fare. However, this can be done in a few subtly different ways.

1. Intentional under-purchasing Distance Travelled:

   A passenger is travelling from Station A to Station D (4 zones). They purchase a ticket to Station B (2 zones) but ride the entire way to Station D anyway.

2. Ungated Stations:

   A passenger is travelling from Station A (an ungated regional station) to Station D (gated city station). They board at Station A with no ticket and while travelling purchase a mobile ticket for the shorter distance of Station C to Station D that will allow them to exit.

3. Split Ticket Scenarios:

   A rider is taking the train from Station A to Station D. Instead of buying a single ticket for the full journey, they buy two cheaper tickets.
   • Ticket 1: Station A → Station B 
   • Ticket 2: Station C → Station D

They tap in at Station A and tap out at Station D but never validate at the middle stations. The ticketing system sees two legitimate trips – but the rider has paid for less than half of what they actually travelled.

Short ticketing is most prevalent in areas where fares are calculated by distance or by zone, as the incentive for taking shorter journeys is higher. Short ticketing practices also represent a key form of fare evasion on barcode-based ticketing systems, where riders can easily purchase a mobile ticket while travelling aboard their journey.

Why Traditional Inspections Don’t Work

Unlike jumping a barrier and riding without a ticket, short ticketing is far more difficult for transit agencies to detect. Transit agencies have long relied on a combination of physical barriers (gates, transit vehicle drivers) and random ticket checks or manual audits to combat fare evasion. But when your rider has a ticket, it becomes a much tougher game of cat and mouse that is very dependent on inspectors being in the right place at the right time.

This issue is compounded by:

• Gate limitations:

  Most fare gates only check if a ticket exists, not whether it covers the entire journey.

• Data gaps:

  Stations use different ID formats; some rider entry/exist events go unrecorded and clever riders can exploit these gaps.

• Complex behaviors

  : While contactless cards, mobile tickets and special passes add convenience to riders, it adds noise to the data.

• Visual inspection:

  Relies on inspectors conducting their inspection during the portion of the trip where the rider isn’t ‘valid.’ 

Finding these under-the-radar patterns is like looking for a mismatched puzzle piece in six million records.

Our Mission: A Smarter, Data-Driven Solution

Working alongside experts from Imperial’s Artificial Intelligence and Data Analytics (AIDA) Lab, we set out to explore how AI could help us detect possible short ticketing hotspots across the network automatically and inform the proactive allocation of resources.

The Breakthrough: A/B/C/D Station Classification

Our first breakthrough came when we started rethinking how stations are described. Instead of only tracking entries and exits, we leveraged machine learning models to classify every station interaction into four roles:

• A – Actual Entry: Where passengers physically start their trip
• B – Declared Destination: The destination printed on their ticket
• C – Declared Origin: Their Ticket’s stated starting point 
• D – Actual Exit: Where passengers physically leave

By comparing a rider’s declared intentions (B & C) with their actual movements (A & D), we could see a story that didn’t quite add up – a key indicator of short ticketing.

Four AI “Detectives” Working Together

Our second breakthrough was the creation of an unsupervised, multi-expert AI System.

We combined four different anomaly detection methods, each with its own strengths:

1. Isolation Forest - Spots global outliers that don’t fit into network-wide patterns.
2. Local Outlier Factor (LOF) – Catches stations acting strangely compared to similar neighbors.
3. One-Class SVM – Flags never before seen fraud patterns. 
4. Mahalanbois Distance – Measures how far data points deviate statistically from the norm.

Think of these as our four detectives, each looking for clues in their own unique way and then comparing notes. We used an adaptive weighting system to ensure that if two algorithms agreed too often, they didn’t drown out other unique insights.

Mapping Fraud Patterns

Using the combined analysis of our AI detectives we were able to identify five distinct fraud behaviors:

• Ghost station - Stations with lots of entry-only or exit only taps.
• Black-Hole Station – Stations with disproportionate exists and missing entry validations.
• Fake Origins - Stations commonly used to fake starting points.
• Micro-tap – Smaller but persistent anomalies
• Function-Loss – Abnormal entry/exit loops

This pattern identification is key to aiding agencies in tailoring short ticketing countermeasures – from more targeted inspections to smarter gate logic – station by station.

What We Found

Understanding these fraud behaviors and patterns to identify thirty (anonymized) high-risk stations across the network. We were also able to develop insights into which stations were at highest risk for which types of short ticketing practices. For example, one airport station was a high target for ghost station behavior with 63% of tickets being entry only, and 37% being exist only. Another Downtown station displayed strong black hole patterns, with 62% exit-only tickets and a 24% entry/exit imbalance.

Armed with these insights, agencies can deploy ticket inspectors and revenue protection staff exactly where they’re needed, potentially recovering millions in lost revenue, while avoiding costly and invasive blanket inspections. Because the system only uses anonymized operational data, it safeguards passenger privacy while delivering actionable insights to agencies.

Why It Matters – For Agencies and Passengers

This study wasn’t just about stopping fare cheats. Preventing Short ticketing is key to ensuring the financial health of our transit agencies and service improvements for everyone. Every pound lost to short ticketing is a pound not reinvested in service improvements. 

Identifying and cracking down on short ticketing also ensures fairness and prevents honest passengers from subsidizing other riders’ fraud. Unfortunately, opportunistic fare evasion can be contagious. If people, see others blatantly cheating the system with no fear of consequences the behavior can spread as individuals think “if they aren’t paying, why should I?”

Preventing short ticketing can lead to better planning and better services for all. Short ticketing leads operational distortions where misreported ridership data can affect service planning and funding. Eliminating short ticketing and increasing the accuracy of ridership data, can lead to smarter schedules and more targeted upgrades.

What’s Next

By pairing Cubic’s transit expertise with the Imperial team’s advanced machine learning capabilities, we were able to show that short ticketing – a problem once nearly invisible – can be detected and mitigated.

While this study relied on a small 100-station dataset, the outputs shone a light on what is possible with the use of AI. We hope to expand the model beyond this limited dataset and look to integrate barcode ticket data and gate-status logs to drive greater accuracy. We are also exploring real-time deployment so agencies can begin to intervene as anomalies happen. We also see great potential for global adoption – from north American commuter rail – to high-speed metros in Asia.

But this is about more than just technology.

It’s about creating fair, financially sustainable public transport systems where everyone pays their share and enjoys the rewards of better service.

Download the full white paper, delivered via Imperial Consultants, for a deeper dive into the algorithms and case studies.