Talking Trash About Smart Waste Management

Waste management is a growing problem in urban areas around the world, with mounting piles of trash and inadequate infrastructure to deal with it. According to the World Bank, the amount of waste generated globally is expected to increase by as much as 70% by 2050, with much of this growth coming from cities in developing countries.

The impact of poor waste management on human wellbeing is significant. According to the World Health Organization, exposure to uncollected waste can lead to a range of health problems, including respiratory and gastrointestinal illnesses. Children are particularly vulnerable to the effects of waste, as they are more likely to play in contaminated areas and are less able to protect themselves from exposure. In addition, uncollected waste can pollute waterways and harm wildlife.

The waste management challenges that exist today are only going to get worse in the future as populations swell in urban areas. Recognizing that this problem is a ticking time bomb, teams of researchers at the University of South Australia and Central Queensland University came together to look for a solution. They believed that the complexities associated with waste management, like developing clearance schedules and optimizing the placement of trash cans, could be better understood through machine learning. The pattern detecting capabilities of this technology should be able to identify actionable trends, they reasoned.

By leveraging sensorized smart trash cans and analyzing the data with a deep learning algorithm, they found that it was possible to get a handle on which areas need frequent trash pickups, and which are rarely used. These insights can also be used to shuffle the locations of trash cans so that they are exactly where they are most needed.

The researchers worked with the City of Wyndham in the suburbs of Melbourne, Australia throughout their research. The city gave them access to data from their Internet-connected smart trash cans that regularly report data such as their geographical coordinates, the current time, and the amount of trash that they contain.

A total of four types of deep learning models were evaluated (1D CNN, LSTM, GRU, and Bi-LSTM) for their ability to predict future waste generation patterns. A full year’s worth of sensor data was acquired and split into training and testing datasets. Each of the four models was then trained using this example data. Similar hyperparameters and optimizers were used in the course of training each model to make each option comparable.

It was revealed that the best performing model was the LSTM option after calculating the train and test error metrics. Given the observed level of error, it was shown that LSTM models can be used for trash can emptiness or fullness prediction with a high degree of accuracy. The insights derived from these predictions can assist planners in developing optimal waste management strategies.

Next up, the team is planning to investigate how organizational and socioeconomic factors, and public utility investment, may influence waste generation. They believe that this work will further assist planners in building an optimized and cost-efficient waste management system.