Machine learning maps which Ghanaian research fields are bleeding talent

Ghana is losing more researchers who publish scientific work than it is gaining. These losses are concentrated in fields that a developing country needs to build its knowledge and skills. A new study published in the journal Discover Data examined 14 years of publication records from Scopus and the Web of Science. The researchers tracked where Ghanaian research talent moved, grouped these movements into categories, and used machine learning to predict future trends. Their results show that more research talent has been leaving Ghana than entering the country, and this gap grew sharply after 2012.

The research team was led by Muftawu Hussein, with colleagues including Emmanuel Ahene and Issifu Damba Kanzoni. They collected 51,000 publications from 2007 to 2020: 29,000 from Scopus and 22,000 from the Web of Science Core Collection. They removed 12,240 duplicate records found in both databases. A pair of records counted as duplicates if they shared the same DOI and matched in at least two other details, such as the first author's name, article title, ISSN, or publication year. This left 38,760 publication records for analysis. The team studied researchers linked to Ghana who had published at least two works during this period. Each record showing a person's authorship was treated as a data point connecting that person to an institution and a country.

The researchers chose this period for specific reasons. Around 2007, many African universities began using standard digital identifiers, such as ORCID and Scopus Author ID. These identifiers made it easier to tell researchers apart and track their connections to institutions and countries over time. The team ended the study period in 2020 to avoid changes caused by the COVID-19 pandemic, which disrupted academic travel, research funding, and publishing around the world. Looking at 14 years of records also gave the researchers enough information to identify major changes in researchers' career paths.

The study grouped researchers into four categories based on the country listed on their first and most recent publications. Emigrants were researchers whose academic careers began in Ghana and whose final listed affiliation was abroad, as well as those who moved between two countries outside Ghana. Immigrants were researchers who began outside Ghana and later published under a Ghanaian affiliation. Return migrants started their careers in Ghana, left the country, and then came back. Non-mobile researchers had affiliations in Ghana throughout the study period. The results were striking. Of the 38,760 researchers identified, 58.1 percent, or 22,519, showed signs of international migration, while 41.9 percent, or 16,240, showed no such movement. Among the mobile researchers, 45.0 percent, or 17,442, moved out of Ghana, while only 13.1 percent, or 5,077, moved into the country. Of those entering Ghana, 5.8 percent had started their careers abroad, while 7.3 percent were researchers who had begun in Ghana, left, and later returned.

Ghana's net research migration rate became negative in 2012 and stayed negative through 2020. From 2007 to 2011, the country gained more researchers than it lost. The biggest net gain came in 2009, at 26.7 researchers per 1,000. Health Sciences had the strongest gain that year, at 13.8 per 1,000, followed by Life Sciences at 6.0, Social Sciences at 4.1, Physical Sciences at 2.0, and multidisciplinary research at 0.8. From 2012 onward, the pattern reversed. By 2019 and 2020, the rate had dropped below minus 30.0 researchers per 1,000. The year 2020 had the largest outflow in the study period. The authors interpret these results as evidence that Ghana was losing substantial research talent, especially in the later years covered by the data.

Looking at individual fields gave a clearer picture of where the losses were greatest. The researchers measured support, which means the share of records that included a particular field. Among emigrating researchers, Medicine had the largest share, with 4,483 scholars, or 25.7 percent. Agricultural and Biological Sciences followed with 2,581 scholars, or 14.8 percent, and Social Sciences had 1,936, or 11.1 percent. Environmental Sciences, Computer Science, Biochemistry, Genetics and Molecular Biology, Business, Management and Accounting, and Engineering also each passed the 3 percent support threshold. When fields were grouped into broader subject areas, Health Sciences had the largest share of losses at 28.9 percent, followed by Life Sciences at 24.1 percent. However, these same fields also attracted researchers to Ghana. Of the 5,077 researchers who returned or moved into Ghana, 33.9 percent, or 1,721, worked in Medicine, while 19.5 percent, or 990, worked in Agricultural and Biological Sciences. The authors describe this as a reverse brain gain happening alongside an overall brain drain.

The study's main method was association rule mining, a data analysis technique often used to find patterns in shopping baskets. The researchers adapted it to publication records. Each data group contained one author's full publication history from 2007 to 2020, including changes in country affiliation and research field. They used the Apriori algorithm to find combinations that appeared often and to create strong rules that passed minimum levels for support and confidence. Each rule was measured using four standard metrics. Support measures how often a combination appears. Confidence measures how likely the result is when the starting condition is present. Lift compares how often two things appear together with how often they would be expected to appear together by chance if they were independent. Conviction measures how strongly a rule avoids cases where its prediction would be wrong under the assumption of independence. A lift above 1 indicates a positive relationship between a research field and a destination, while a lift below 1 indicates a negative relationship.

The rules showed where Ghanaian researchers were moving. The rule linking Medicine to the United States had a confidence of 0.439, meaning there was a 43.9 percent likelihood that Ghanaian medical scholars would migrate to the United States under the rule. Its lift was 1.288 and its conviction was 1.175, indicating a moderate positive relationship. The rule linking Medicine to the United Kingdom had a confidence of just 12.6 percent, although the United Kingdom was the only other destination with a significant probability for this field. For Computer Science, the confidence of a link to the United States was 39.4 percent. For Social Sciences, it was 38.0 percent. Their lift scores were 1.154 and 1.115, respectively. One result showed the opposite pattern: Agricultural and Biological Sciences had a lift of 0.572 and a conviction of 0.818 for movement to the United States, indicating a negative relationship. Among researchers moving into Ghana, the strongest link was between Biochemistry, Genetics and Molecular Biology and the United States. This rule had a lift of 2.729, a conviction of 2.056, and a confidence of 62.5 percent. It suggests that many researchers in these fields began their careers in the United States before later appearing in Ghana's publication records. Across emigrants, immigrants, and return migrants, the United States and the United Kingdom were the most prominent countries of origin, transit, and destination. China, South Africa, Germany, Canada, and Nigeria also appeared repeatedly.

To predict future trends, the team used time-series modeling in PyCaret, a machine learning tool that requires relatively little coding. They focused on the two fields with the highest emigration rates: Medicine and Agricultural and Biological Sciences. The models used only the year and field of study as input features. The researchers divided the data into 65 percent for training and 35 percent for testing. The test years came after the training period, keeping the data in time order. They trained and assessed the models using fivefold cross-validation, a method that tests model performance across five different data splits. The best model achieved a mean absolute error of 0.04893, with a root mean square error of a similar size. These measures describe the size of the model's prediction errors. The forecasts, which extended toward 2030, suggest that both fields will continue to face pressure from researchers leaving Ghana. The results support the study's conclusion that Social Sciences, Agricultural and Biological Sciences, and Health Sciences are especially at risk of losing research talent over the long term.

The authors explain the findings using three ideas: push–pull theory, human capital theory, and network theory. Poorly funded laboratories and limited opportunities for career growth can push researchers to leave. Better facilities, stronger funding, and greater recognition can attract them to established research systems. Networks of researchers living abroad and international research partnerships can also guide where people move, creating routes that remain active over time. The authors argue that predictions like these can give policymakers early warnings and help them develop focused plans to keep researchers in Ghana, work with researchers living abroad, and encourage those who left to return. However, the study has important limits. Publication databases miss researchers who publish in local journals or journals that are not included in the databases. Matching authors by their names and affiliations cannot guarantee that every researcher is identified correctly. The models also do not investigate the personal, workplace, or financial reasons behind individual moves, and they cannot predict sudden events such as policy changes or global crises. The authors note that the same method can be used in other countries that want to understand where their research talent is going.

Subject of Research: Predictive analysis of international migration patterns among Ghanaian researchers using bibliometric data and machine learning

Article Title: Robust predictive analysis of international mobility among research talents

Article References: Hussein, M., Ahene, E., Hassan, A. L., & Kanzoni, I. D. (2026). Robust predictive analysis of international mobility among research talents. Discover Data, 4(1), Article 16. https://doi.org/10.1007/s44248-026-00110-5

Image Credits: AI-generated

DOI: 10.1007/s44248-026-00110-5

Keywords: Ghana, brain drain, researcher mobility, bibliometrics, association rule mining, machine learning, Scopus, Web of Science, net migration rate, talent retention, health sciences, time series forecasting