Information and evidence in Yemen: a case for using predictive analytics and cross-sector collaboration for anticipatory action

Embroiled in civil war since 2015, Yemen has become one of the world’s worst humanitarian crises. This conflict, rooted in the failed Arab Spring uprising in 2011, is exacerbating several other issues. Already the poorest country of the region, Yemen’s conflict has destroyed its economy, infrastructure and livelihoods.

It has also driven displacement, food insecurity and extreme poverty. According to the United Nations, in 2021 an estimated 24.3 million people were at risk of hunger and disease, of whom 14.4 million were considered in acute need of assistance. Furthermore, 20.5 million people are without access to safe drinking water and sanitation, and 16.2 million people require emergency assistance as a result of malnutrition and food insecurity. In return, food scarcity has led to the outbreak of diseases such as cholera and dengue.

Anticipatory action using predictive analytics in humanitarian contexts can range from prediction models of flood forecasts, mapping hunger trends, and early warning for disease outbreaks such as cholera – all of which can be used to save countless lives. These humanitarian actions, taken in advance of a hazard, require reliable information and evidence, but obtaining these during a crisis often sees many barriers and challenges related to the quality and quantity of data. These include accessing data, missing and biased data, and a lack of transparency, verification and data sharing. In Yemen, for example, data on food security, malnutrition and famine is controlled by the internationally recognized government of Aden or the Houthis in Sana’a, which results in misreporting of information. Routine checks and verification of data are prohibited or limited, which brings its quality, independence and neutrality into question.

In addition, the data cannot be taken out of the country and there are limits on sharing it externally, limiting its accessibility. There are issues of missing data, outdated data, and data that doesn’t measure the necessary unit of analysis. There are also gaps in geographic coverage, making it difficult to identify ‘hotspots.’

Overcoming challenges through better data

Given the nature of data challenges, in Yemen and elsewhere, some recommendations should be explored. Remote sensing technology, open-source data and predictive analytics can and should be used to compensate for the lack of primary data. Predictive analytics incorporates the use of machine learning (ML) methods and artificial intelligence (AI) to create humanitarian ‘forecasts’ and ‘predictions’. These tools utilize various forms of data, including historic data, spatial data and remote-sensing data, to build predictive models using appropriate algorithms. The caveat with ML and AI is that these models are only as good as the data being used, and there is a threat of biased analysis which can result in inaccurate predictions. However, with more information, these models can be trained and modified to present more accurate and realistic scenarios.

AI is already playing a role in humanitarian action. The Water, Peace and Security global early warning tool identifies emerging water-related conflicts up to one year in advance, which then guides interventions ahead of a crisis. The Famine Early Warning Systems tool uses mixed methods as an early warning tool for food insecurity. Many other humanitarian programmes have also created tools to estimate costs, and to model and forecast conflicts, displacement and natural disasters (e.g., floods and diseases).

Another benefit of ML models and AI is that they can be employed remotely. During the COVID-19 pandemic, the humanitarian sector was largely forced to operate remotely, which imposed an additional barrier to data collection. But ML and AI tools can supplement an inability to collect data on the ground. Remote-sensing technologies and open-source data are also becoming increasingly affordable and available.

The need for cross-sector collaboration

One of the biggest issues in evidence-gathering and information-sharing is that organizations tend to collect data and work in silos. Therefore, humanitarian actors across sectors and agencies should collaborate, share and consolidate data. Supporting cross-sector collaboration is critical to gain a more holistic understanding of crisis situations, including in Yemen.

In recent years, there has been an expansion of publicly available datasets, which now cover such diverse themes as agricultural production, land cover and land use, prices, climate, conflict, mobile phone use and social media. Many are relevant to humanitarian work, such as the Armed Conflict Location & Event database, the Uppsala Conflict Data Program and the Yemen Data Project.

Social media can also inform humanitarian actors about the conditions on the ground. Although internet access can be scarce in crisis contexts, social media platforms such as Twitter, Facebook and YouTube can, when available, provide useful information that can be mapped. This can support humanitarian actors to identify the location and nature of events occurring in their region of operation.

To address natural hazards, there are many open-source climate datasets and remote-sensing technologies that can be used to build predictive models – an essential element of anticipatory action. Examples include the World Resource Institute’s Aqueduct initiative and the IPCC Working Group’s Interactive Atlas. In Yemen, NASA’s Earth Observatory used data on precipitation, air temperature and population to forecast the risk of a cholera outbreak. Their model ultimately proved to have a 92 per cent accuracy in its predictions. Acting ahead of disease outbreaks is a growing field of interest, and AI and ML should play a prominent role in advancing this.

All humanitarian crises have their own challenges with obtaining transparent, neutral, independent, high-quality and reliable data – and acting in anticipation of a crisis is no different. AI and ML applications, remote-sensing technologies, open-source data and predictive analytics can supplement the lack of data in humanitarian contexts and play a central role in overcoming these data challenges. This should be done alongside increased cross-sector and cross-agency collaboration and data-sharing to develop a more complete picture of the situation on the ground.