Despite the dizzying complexity of all its components, your body is still extraordinarily efficient at what it does. But what if it wasn’t? What if your brain had to manually and non-intuitively chart tedious plans to coordinate what blood flows through which arteries at what time?
Termed by some as the arteries of our economic system, India boasts the fourth largest railway network worldwide, a technological marvel whose function propels a flurry of crucial socioeconomic development across the country’s various corridors. But how do you upgrade anything without painstakingly replacing hardware that has stood firmly on the nation’s soil for over 160 years? According to a team of collaborators from Zonal Railways, the Centre for Railway Information Systems (CRIS), and the Indian Institute of Technology Bombay (IIT Bombay), you don’t have to — you just need to look at things with fresh eyes.
To understand this new research involving Prof Madhu Belur from the Department of Electrical Engineering, Prof Narayan Rangaraj from the Department of Industrial Engineering and Operations Research at IIT Bombay, and experts from Zonal Railways and CRIS, let’s first look at how trains are timetabled.
In this study, the team grouped Indian trains into two broad categories: daily trains — trains that run on all days of the week — and non-daily trains that run a few times per week. For example, most long-standing trains between popular destinations run daily. In contrast, some trains to very specific destinations run only on specific days of the week — this is quite common when new trains are introduced between pairs of cities where traffic is building up over time. While daily trains have it mostly sorted out, the non-daily ones are a tougher challenge within the Indian railway timetable.
The main issue is that these non-daily trains are scattered throughout the week, making it hard to plan efficient schedules. Another challenge is that different railway zones initially plan their local schedules considering their own sectional resources, and this may result in these non-daily trains having conflicts at other busy sections or sub-optimal use of bottleneck resources on the larger network. To address this, the team of collaborators explored a process called ‘dailyzing,’ which involves clustering non-daily trains to improve scheduling efficiency.
At its core, dailyzing is about grouping “similar” non-daily trains into a single, predictable pattern, almost as if they were a daily service. Similar trains that use almost the same resources and at almost the same times (within a window of 15 min) but on different days of the week are grouped in a cluster. Instead of treating these sporadic trains as independent, railway planners can now map them onto a 24-hour schedule, creating a structured timetable that fills gaps and reduces inefficiencies.
The team used Hierarchical Agglomerative Clustering (HAC) — a technique that identifies patterns in vast amounts of data — to sort trains that run on similar routes at similar times but on different days. These trains are then scheduled as a single "cluster", ensuring a more compact and efficient timetable. The team found that scheduling one representative train from the group as a ‘daily train’ can help to schedule non-daily trains in that group.
Right now, the Indian Railways operates over 13,150 passenger trains daily across the country, yet many of its non-daily services run inconsistently across the week. This scattered scheduling leads to underutilised tracks on some days and congested bottlenecks on others. By clustering trains, the researchers found they could speed up the timetabling process because once a cluster is scheduled, every train within it automatically follows suit.
Think of it like scheduling a bus for a busy city junction. If five different buses pass through the same junction at the same time of the day but on different days of the week, it would be rather tiresome and inefficient to schedule it as such. But when you cluster all the buses into one “daily” path, you just have to plan for one bus, which will automatically set the times for all the buses in that group.
Further, this method could even make space for new trains. If a cluster has fewer than seven trains (one train for each day of the week), additional services can be slotted in the free days, the research explained. Bottleneck sections could now be managed more effectively, reducing delays and maximising train flow.
To put their model to the test, the researchers focused on India's Golden Quadrilateral and Diagonals (GQD) network — a vast rail system connecting Delhi, Mumbai, Chennai, and Kolkata. They analysed real-world train data and applied popular clustering techniques such as Hierarchical Agglomerative Clustering (HAC), Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and K-means. The results were striking, showing that HAC produced the best clusters, ensuring that non-daily trains mostly “complemented” each other, rarely causing conflicts. In this context, a cluster satisfies the complementing conditions once all its trains run on the same stretches and at the same time of the day but on different days of the week.
HAC produced conflict-free clusters within seconds, while some other techniques often took several minutes. This speedy method using HAC also helped reveal hidden inefficiencies, pointing out where new trains could be introduced without disrupting the system.
Prof Madhu Belur explained that the model was based on and built for the GQD network for two reasons: the GQD comprises a significant and major portion of the total freight and coaching traffic for the Indian Railways. Secondly, timetables for non-GQD and less frequented railway zones are carried out in a much more zonal fashion, which already simplifies things enough since the lighter volume of trains in these areas makes it easier to timetable their schedules.
Interestingly, Prof Belur noted that Indian Railways, with the team’s support in creating an automated tool based on clustering, has already been implementing a modified version of the dailyzing model to enhance timetabling on the GQD. Future refinements could further enhance train scheduling by fine-tuning clusters to include more trains and integrating real-time adjustments, making India's railway system even more seamless and adaptable.
The researchers acknowledge that as the model scales up, new challenges will emerge. Long-distance trains often pass through multiple congested sections, where a single clustering approach may not be enough. A train that fits neatly into a cluster in one section of its journey may face bottlenecks elsewhere, requiring a different clustering strategy for different segments of the route. Moreover, since railway zones currently handle their timetables independently, future timetabling refinements will need better coordination across zones to fully realise the benefits of dailyzing. Addressing these complexities will be crucial in making Indian Railways even more efficient and adaptable in the years to come.
This article has been edited to fix a technical error while publishing. The error is regretted