HomeMachine LearningMachine Learning NewsCreating Safe drinking water using Machine Learning

Creating Safe drinking water using Machine Learning

Water-borne diseases are one of the main causes of infectious disease outbreaks in refugee and internally displaced person (IDP) settlements, but a team led by the University of York has developed a new technology that can use machine learning to make drinking water safe. Probably a game changer. The study is published in the journal PLOS Water. Since drinking water is not piped into homes in most settlements, residents have turned to storage containers to collect drinking water from public taps.

When water is stored in residential spaces, it is susceptible to contaminants, so there must be enough residual free chlorine to kill any pathogens, says Lassonde School of Engineering Ph.D. student Michael De Santi is a member of York University’s Dahdaleh Institute for Global Health and led the study.

In refugee and IDP settlements in Kenya, Malawi, Sudan, South Sudan and Uganda, previously safe drinking water was re-contaminated with cholera, hepatitis E and shigellosis during collection, transport and storage, which was a major factor in the outbreak.

Several factors can affect the decay of chlorine in stored water. At that collection point, you can get clean water, but after you take it home and store it, sometimes for 24 hours, you can lose residual chlorine, pathogens can multiply and diseases can spread,” Lassonde Adjunct Professor Imran Ali told researcher Dahdaleh of Global Health institute in York, has first-hand experience of working in a South Sudan settlement. Using machine learning, a team of researchers, including Associate Professor Usman Khan, also from Lassonde, developed a new method to predict the probability that there will be enough chlorine before the last glass of chlorine is consumed. They use Artificial Neural Networks (ANN) and Ensemble Prediction Systems (EFS), which is not commonly done. EFS is a probabilistic model commonly used to predict the probability of precipitation in weather forecasts.

ANN-EFS can generate point-of-use predictions that take into account various factors affecting residual chlorine content, unlike the typical models used. This new probabilistic modeling replaces the universal guidelines currently used for the use of chlorine, which have proved ineffective, ”said Ali.

Unlike commonly used models, ANN-EFS can generate point-of-use predictions that take into account various factors affecting residual chlorine levels. This new probabilistic model replaces the current general guidelines for chlorine use, which have been shown to be ineffective,” Ali said. Factors such as local temperature, water storage and treatment between households, type and quality of water pipes, water quality and whether children put their hands in water containers can affect the safety of drinking water.

However, it is important that these probabilistic models are trained using data from specific populations, as each model is as unique as a snowflake,” said De Santi. “Two people can collect the same water on the same day, both keep it for six hours, and one may still have all the chlorine left in the water, and the other may have almost none. Another 10 people may have a different choice chlorine.

The researchers used regular monitoring of water quality data collected through the Safe Water Optimization Tool project from two refugee settlements in Bangladesh and Tanzania. In Bangladesh, MSF collected data from 2,130 samples from Camp 1, Kutupalong-Balukali Extension Point, Cox Bazaar between June 2019 and December 2019, where the camp hosted 83,000 Rohingya refugees from neighboring Myanmar. Determining how to teach ANN-EFS to make realistic probabilistic predictions with the smallest possible error requires thinking outside the box.

Measuring the error is key because it determines how the model behaves in the context of probabilistic modeling,” De Santi said. “Using cost-aware learning, a tool that transforms cost functions into objective actions using machine learning, we found that it improves probability predictions and reliability. In this case, we don’t know if anyone has done this before.

For example, the model may say that under certain conditions at the tap, where the water contains a certain amount of residual free chlorine, there is a 90% chance that the residual chlorine in the stored water will be below the safe level after 15 hours for drinking.

That’s the probabilistic confidence that the model can give us,” DeSanti said. “Just like a weather forecast, if there’s a 90 percent chance it’s going to rain, you should bring an umbrella. Instead of using umbrellas, we can ask water operators to increase chlorine levels so more people have safe drinking water. ”

Our safe water optimization tool uses this machine learning to work and can help field workers. The only difference for water operators is that we ask them to take multiple water samples in a container at the tap and the same container at home. Hours later, Ali said. “The work Michael has done is the development of the next generation of machine learning models. It can be used not only for providing safe drinking water in refugee and IDP settlements, but also for other applications.

Source link

 

- Advertisment -

Most Popular

- Advertisment -