Municipal solid waste management forecasting using neural networks at discharge point scale

Smarter Trash Pickup for Crowded Cities

As cities grow, so does their trash. Overflowing bins, noisy trucks, and higher greenhouse gas emissions are everyday signs that waste collection is struggling to keep up. This study shows how artificial intelligence (AI) can help cities move from fixed, often wasteful collection schedules to smarter, data-driven planning that knows when and where bins are actually filling up.

Why the Smallest Pieces of the System Matter

Most current waste planning looks at big, citywide numbers: total tons collected each day or month. That helps with long-term budgeting but ignores what really frustrates residents—an overflowing bin on their corner while another is half empty. The authors argue that the key unit to watch is the “discharge point,” meaning each individual bin, container, or inlet where people drop off their trash. By treating each of these points as its own sensor of local behavior, city services can understand how waste builds up street by street and hour by hour, rather than just in yearly totals.

Turning Raw Bin Activity into Usable Signals

To test this idea, the researchers worked with real data from 200 discharge points in a small city in northern Spain, recorded over four years. Every time a bin filled or was emptied, a record was created, producing millions of individual events. This raw stream was messy: readings came at irregular times, some bins had far more measurements than others, and occasional errors or abrupt emptying events distorted the data. The team first cleaned and reshaped these records, aligning them into regular five-minute, then hourly, time steps and adding simple calendar information such as time of day, day of week, month, season, weekend, or holiday. They also detected and smoothed out strange spikes and used a common scaling method (Min–Max normalisation) in some tests to see whether putting all values on the same scale helped.

Training Neural Networks to Anticipate Bin Filling

With this structured dataset, the authors trained artificial neural networks—flexible AI models well suited to capturing complicated, non-linear patterns. The goal was to predict how much waste would be added to a bin over the next few hours. They tested three main strategies. In the first, each discharge point got its own model tailored to its local behavior. In the second, data from several bins were combined into groups, using averages or medians to represent shared patterns. In the third, a single “global” model tried to learn from all bins at once. They also experimented with different time windows, from one to six hours, and with and without data normalisation.

What Worked Best for Reliable Forecasts

The results showed that context matters: adding simple time-related clues to the basic fill readings made a big difference. Models using only a single input field rarely reached accuracy levels that would be useful in practice. When the neural networks combined hourly fill increments with calendar and holiday information, predictions improved sharply. The most reliable forecasts came from grouped-bin models at four- and six-hour intervals, which smoothed out random ups and downs while still capturing daily rhythms. These setups reached accuracy levels that the authors consider strong enough to guide real operations, whereas the one-size-fits-all global model fell short. Interestingly, normalising the data was not always helpful; in several cases, unscaled values gave slightly better forecasts.

From Reactive Cleanup to Proactive Planning

In plain terms, the study shows that cities can use existing bin-level data, combined with simple calendar facts, to predict when and where trash will pile up in the next few hours. Instead of driving fixed routes or responding only after overflow occurs, waste trucks could be dispatched where they are actually needed, at the right time of day. That means fewer unnecessary trips, lower fuel use and emissions, and cleaner streets. While the work focuses on one city and a limited number of bins, it lays out a clear, reusable framework for turning scattered operational records into an AI-powered planning tool—an important step toward more sustainable, responsive urban waste management.

Citation: De-la-Mata-Moratilla, S., Gutierrez-Martinez, JM. & Castillo-Martinez, A. Municipal solid waste management forecasting using neural networks at discharge point scale. Sci Rep 16, 6903 (2026). https://doi.org/10.1038/s41598-026-38110-9

Keywords: municipal solid waste, smart cities, neural networks, waste collection forecasting, urban sustainability