{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,2,27]],"date-time":"2026-02-27T05:11:22Z","timestamp":1772169082155,"version":"3.50.1"},"update-to":[{"DOI":"10.12688\/gatesopenres.14991.1","type":"new_version","label":"New version","source":"publisher","updated":{"date-parts":[[2024,11,20]],"date-time":"2024-11-20T00:00:00Z","timestamp":1732060800000}}],"reference-count":7,"publisher":"F1000 Research Ltd","license":[{"start":{"date-parts":[[2024,11,20]],"date-time":"2024-11-20T00:00:00Z","timestamp":1732060800000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000865","name":"Bill and Melinda Gates Foundation","doi-asserted-by":"publisher","award":["INV-005809"],"award-info":[{"award-number":["INV-005809"]}],"id":[{"id":"10.13039\/100000865","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["gatesopenresearch.org"],"crossmark-restriction":false},"short-container-title":["Gates Open Res"],"abstract":"Background Tobacco use trends among adolescents in low- and middle-income countries, and in particular narrowing gender gaps, highlight the need for interventions to prevent and\/or reduce tobacco use among adolescent girls. We evaluated a social marketing program in Ghana discouraging tobacco use among adolescent girls and additionally investigated the pathways influencing smoking behaviors to identify programmatic opportunities for impact. Leveraging the data collected through the stepped wedge cluster randomized trial and panel survey of 9000 girls aged 13\u201319 , we sought to apply machine learning (ML) techniques to identify the most important variables for predicting initiation of smoking. Methods To identify predictors of smoking initiation we sought to develop a model which could accurately differentiate smokers from non-smokers and evaluated various ML approaches for training classifier algorithms to achieve this. We selected a Synthetic Minority Over-sampling Technique (SMOTE) because it optimized the recall and precision of the model. We then utilized the technique of feature importance for greater insight into how the model arrived at its decisions and to rank the most important variables for predicting smokers. To explore different dimensions of smoking behavior, including initiation and continuation, we trained our model by using several combinations of target outcomes and input variables from the panel survey. Results The resulting features of smokers highlight the importance of girls\u2019 independence and connectivity, social environment, and peer influence on likelihood of smoking, and in particular subsequent initiation. These results were largely consistent with our formative research findings based on qualitative interviews informed by behavioral science. Conclusions This novel application of ML techniques demonstrates how data science approaches can generate new programmatic insights from rigorous evaluation data, especially when data collection is informed by behavioral theory. Such insights about the relative importance of different features can be valuable input for program planning and outreach.<\/ns3:p>","DOI":"10.12688\/gatesopenres.14991.2","type":"journal-article","created":{"date-parts":[[2024,11,20]],"date-time":"2024-11-20T10:05:10Z","timestamp":1732097110000},"page":"2","update-policy":"https:\/\/doi.org\/10.12688\/gatesopenres.crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Application of machine learning techniques to profile smoking behavior of adolescent girls in Ghana"],"prefix":"10.12688","volume":"8","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-1707-608X","authenticated-orcid":false,"given":"Sara V.","family":"Flanagan","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Ariadna","family":"Vargas","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6400-8050","authenticated-orcid":false,"given":"Jana","family":"Smith","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"2560","published-online":{"date-parts":[[2024,11,20]]},"reference":[{"key":"ref-1","doi-asserted-by":"publisher","first-page":"255-260","DOI":"10.1126\/science.aaa8415","article-title":"Machine learning: trends, perspectives, and prospects.","volume":"349","author":"M Jordan","year":"2015","journal-title":"Science."},{"key":"ref-2","doi-asserted-by":"publisher","DOI":"10.1186\/s12889-024-20413-z","article-title":"Social norms and social opportunities: a qualitative study of influences on tobacco use among urban adolescent girls in Ghana.","volume":"24","author":"L Aryee","year":"2024","journal-title":"BMC Public Health."},{"key":"ref-3","first-page":"559-563","article-title":"Imbalanced-learn: a Python toolbox to tackle the curse of imbalanced datasets in machine learning.","volume":"18","author":"G Lemaitre","year":"2017","journal-title":"J Mach Learn Res."},{"key":"ref-4","doi-asserted-by":"publisher","first-page":"1322-1328","DOI":"10.1109\/IJCNN.2008.4633969","article-title":"ADASYN: adaptive synthetic sampling approach for imbalanced learning.","author":"H He","year":"2008"},{"key":"ref-5","doi-asserted-by":"publisher","first-page":"321-357","DOI":"10.1613\/jair.953","article-title":"SMOTE: Synthetic Minority Over-sampling Technique.","volume":"16","author":"N Chawla","year":"2002","journal-title":"J Artif Intell Res."},{"key":"ref-6","doi-asserted-by":"publisher","first-page":"136-158","DOI":"10.1007\/s10618-011-0222-1","article-title":"Hellinger distance decision trees are robust and skew-insensitive.","volume":"24","author":"D Cieslak","year":"2012","journal-title":"Data Min Knowl Disc."},{"key":"ref-8","first-page":"2825-2830","article-title":"Scikit-learn: machine learning in python.","volume":"12","author":"F Pedregosa","journal-title":"J Mach Learn Res."}],"container-title":["Gates Open Research"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2\/xml","content-type":"application\/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2\/iparadigms","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,8,26]],"date-time":"2025-08-26T10:45:06Z","timestamp":1756205106000},"score":1,"resource":{"primary":{"URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2024,11,20]]},"references-count":7,"URL":"https:\/\/doi.org\/10.12688\/gatesopenres.14991.2","relation":{"has-version":[{"id-type":"doi","id":"10.12688\/gatesopenres.14991.1","asserted-by":"subject"}],"has-review":[{"id-type":"doi","id":"10.21956\/gatesopenres.17698.r39728","asserted-by":"subject"},{"id-type":"doi","id":"10.21956\/gatesopenres.17698.r38488","asserted-by":"subject"},{"id-type":"doi","id":"10.21956\/gatesopenres.17698.r38488","asserted-by":"object"},{"id-type":"doi","id":"10.21956\/gatesopenres.17698.r39728","asserted-by":"object"}]},"ISSN":["2572-4754"],"issn-type":[{"value":"2572-4754","type":"electronic"}],"subject":[],"published":{"date-parts":[[2024,11,20]]},"assertion":[{"value":"Indexed","URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2#article-reports","order":0,"name":"referee-status","label":"Referee status","group":{"name":"current-referee-status","label":"Current Referee Status"}},{"value":"10.21956\/gatesopenres.17698.r38488, Ruoyan Sun, The University of Alabama at Birmingham, Birmingham, Alabama, 11 Dec 2024, version 2, indexed","URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2#referee-response-38488","order":0,"name":"referee-response-38488","label":"Referee Report","group":{"name":"article-reports","label":"Article Reports"}},{"value":"10.21956\/gatesopenres.16324.r36854, Jennifer M. Murray, Queens University Belfast, Belfast, England, 17 Jul 2024, version 1, 2 approved with reservations","URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v1#referee-response-36854","order":1,"name":"referee-response-36854","label":"Referee Report","group":{"name":"article-reports","label":"Article Reports"}},{"value":"Sara Flanagan<\/b>; \nPosted: 19 Nov 2024<\/i>; There is little discussion on the current literature on adolescent smoking in the introduction section of the main text. For example, although the background section of the abstract mentions tobacco trends in low- and middle-income countries, and narrowing gender gaps, this is not expanded upon in the main text and there is no mention of the adolescent smoking prevalence (for girls in Ghana). I appreciate that the article is a Research Note and therefore has a limited word count, but I would still expect to see some reference to the issue of adolescent smoking rates in the introduction section. \nResponse<\/b>: This background info has been added to the Introduction section of the main text along with citations. Similarly, throughout the paper the authors highlight that their survey items (and the ML inputs) were informed by the formative\/qualitative research findings, and the behavior change theory underlying the evaluated program. However, the qualitative findings or the behavioral theory have not been summarized anywhere. I think it would be interesting to see a brief overview of the social marketing program, and how it relates to the study's outcome measures, in the methods section. \nResponse<\/b>: The qualitative study has just recently been published and the citation has been updated. That paper includes more explanation of the context for the study and an extensive discussion of the findings relative to the literature. At the end of the methods section, you state that you conducted some secondary analyses using the same data inputs and predicted outcomes within subgroups of the respondents (including younger versus older girls, and different regions), and also used different combinations of data inputs and predicted outcomes. However, you were not very specific about these analyses, and the results are not reported. I suggest you should add specific details about these analyses (e.g., what age groups and regions were compared, what alternative combinations of data inputs and predicted outcomes were used) and summarize the findings in the results section. You could also upload the results tables for the secondary analyses to a repository and provide the link in the "Extended data" section. I do not think this should be left as it is because it is vague, and the results have not been discussed anywhere in the article. It would be interesting to see the results if you had used the "intentions for smoking cigarettes or shisha over the next 30 days" outcome at the end of the survey. \nResponse: <\/b>We have updated the methods to specify the age groups (13-15 vs. 16-19) and regions (Accra vs. Kumasi), and summarized the age subgroup results for model 3 at the end of the results section. We also edited the Methods so that it does not seem like we explored additional outcomes (we did not explore intention to smoke), rather we looked at the same outcomes (ever smoking, recent smoking) at different time points. Similarly, in the methods section you state that you recoded some of your data's Likert scales to binary. This would cause some loss of information. Have you considered presenting alternative (sensitivity) analyses to determine the impact of using the original uncategorized outcome variables? \nResponse: <\/b>SMOTE uses Euclidean distance to figure out how close minority samples are to each other, but can struggle with high-dimensional or ordinal data, like raw Likert scales, because of a problem called \u201cdistance distortion.\u201d In high-dimensional spaces, data points are more spread out, which makes their distances more similar and makes it hard to find true nearest neighbors. This distortion could make a sensitivity analysis confusing, since it will be difficult to know if any differences are from real predictive value or just from high-dimensional effects. Moreover, transforming the Likert scale variables was also to make interpretation of the feature importance technique easier, so using the original scales again may not add much value, given potential distortion, and could reintroduce complexity to the interpretation. Although you have provided a link to the full survey in the "Extended data" section, I think you could provide some more information on the outcome measures in the methods section (e.g., insert the relevant references if they have been adapted from previous research studies, and describe whether they have been validated for use with your research population). \nResponse: <\/b>We added a comment that the tobacco use questions were adapted from the Global Youth Tobacco Survey, and cited the formative research, which is the context in which the questionnaire was tested in this population. The study's strengths and limitations have not been discussed in the discussion section. \nResponse: <\/b>The final paragraph of the discussion mentions the limitations of data inputs and the strengths of grounding data collection in behavioral science. We have expanded there to mention the large sample size as a strength and referenced a longer discussion of limitations related to collecting data from this study population in our formative research paper, to save on space here. On page 4, where you provide an overview of the ML techniques, you state that the "RandomOverSampler" technique works by "randomly oversampling the minority class (non-smokers)". Are smokers not the minority class? \nResponse: <\/b>This was a typo and has been corrected, we thank the reviewer for catching it. Where you describe the SMOTE technique (the optimal technique whose results you ultimately report on), you say that it works by generating samples based on the k-nearest neighbors of the minority samples. What was "k" in your models? \nResponse: <\/b>We used 5 k-neighbors and added this detail to the description of the technique. In your models, "lives in Accra" is frequently an important predictor. Can you comment on this result (e.g., the differences between Accra and the other communities in your sample, and why adolescents living in Accra should have increased risk of smoking)? \nResponse: <\/b>We have added a comment to the results discussion that Accra is the more cosmopolitan and diverse of the two study cities.","URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2#referee-comment-3749","order":2,"name":"referee-comment-3749","label":"Referee Comment","group":{"name":"article-reports","label":"Article Reports"}},{"value":"10.21956\/gatesopenres.17698.r39728, Runqiu Wang, University of Nebraska Medical Center, Omaha, Nebraska, 26 Aug 2025, version 2, indexed","URL":"https:\/\/gatesopenresearch.org\/articles\/8-2\/v2#referee-response-39728","order":3,"name":"referee-response-39728","label":"Referee Report","group":{"name":"article-reports","label":"Article Reports"}},{"value":"This work was supported by the Bill and Melinda Gates Foundation [INV005809].","order":4,"name":"grant-information","label":"Grant Information"},{"value":"This is an open access article distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.","order":0,"name":"copyright-info","label":"Copyright"}]}}