To what extent an individual or group will be affected by the damage of a hazard depends not just on their exposure to the event but on their social vulnerability – that is, how well they are able to anticipate, cope with, resist, and recover from the impact of a hazard. Therefore, for mitigating disaster risk effectively and building a disaster-resilient society to natural hazards, it is essential that policy makers develop an understanding of social vulnerability. This study aims to propose an optimal predictive model that allows decision makers to identify households with high social vulnerability by using a number of easily accessible household variables. In order to develop such a model, we rely on a large dataset comprising a household survey (n = 41 093) that was conducted to generate a social vulnerability index (SoVI) in Istanbul, Türkiye. In this study, we assessed the predictive ability of socio-economic, socio-demographic, and housing conditions on the household-level social vulnerability through machine learning models. We used classification and regression tree (CART), random forest (RF), support vector machine (SVM), naïve Bayes (NB), artificial neural network (ANN), k-nearest neighbours (KNNs), and logistic regression to classify households with respect to their social vulnerability level, which was used as the outcome of these models. Due to the disparity of class size outcome variables, subsampling strategies were applied for dealing with imbalanced data. Among these models, ANN was found to have the optimal predictive performance for discriminating households with low and high social vulnerability when random-majority under sampling was applied (area under the curve (AUC): 0.813). The results from the ANN method indicated that lack of social security, living in a squatter house, and job insecurity were among the most important predictors of social vulnerability to hazards. Additionally, the level of education, the ratio of elderly persons in the household, owning a property, household size, ratio of income earners, and savings of the household were found to be associated with social vulnerability. An open-access R Shiny web application was developed to visually display the performance of machine learning (ML) methods, important variables for the classification of households with high and low social vulnerability, and the spatial distribution of the variables across Istanbul neighbourhoods. The machine learning methodology and the findings that we present in this paper can guide decision makers in identifying social vulnerability effectively and hence let them prioritise actions towards vulnerable groups in terms of needs prior to an event of a hazard.

Amaç: Güç analizi ile örneklem büyüklüğü tahmini tıbbi araştırmacılar ve etik kurulu üyeleri tarafından çoğu zaman yanlış yorumlanan bir konudur. Bu çalışmada, istatistik alanı dışından olan araştırmacıların, örneklem büyüklüğü, etki büyüklüğü ve istatistiksel güç ile ilgili sorularının cevaplanması amaçlanmıştır. Yöntem: Bu kapsamda, farklı tıbbi araştırma düzenlerinde güç analizi gerçekleştirmek için kullanılacak etki büyüklüklerinin ne anlama geldiği ve nasıl hesaplandığı anlatılmıştır. Farklı etki büyüklüklerinde, farklı istatistiksel güç düzeylerinde ve %5 istatistiksel anlamlılık seviyesinde bağımsız gruplar için t-testi, tek yönlü ANOVA ve Ki-kare testleri için gerekli olan örneklem büyüklükleri GPower 3.1 programı kullanılarak hesaplanmıştır. Bulgular: Farklı istatistiksel testler için gerçekleştirilen güç analizleri, etki büyüklüğünün örneklem sayısı belirlemede ne derecede önemli rol oynadığını göstermiştir. Gereğinden az sayıda birim üzerinde deney yapılması gerçekte var olan önemli bir etkinin tespit edilememesine neden olabilirken, çok fazla sayıda örnek üzerinde test yapılması istatistiksel olarak anlamlı fakat gerçekte klinik olarak önemsiz bir etkinin tespit edilmesine yol açabilir. Sonuç: Sonuç olarak, bir araştırma kapsamında beklenen klinik anlamlılığı ortaya çıkarmak için alınması gereken etki büyüklüğü, araştırma hipotezi ile uyumlu belirlenmelidir. Çünkü araştırma sonucunda verilecek kararların niteliği ve klinik anlamlılığı örneklemin hangi etki büyüklüğüne dayanarak seçildiğine bağlıdır.

Machine learning techniques (MLTs) are increasingly being used to develop clinical risk prediction models for binary health outcomes but the sample size requirements for developing and validating such models remain unclear. This study investigates whether sample size guidelines that target mean absolute prediction error (MAPE) for logistic regression models can be applied to tree-based ensemble MLTs (bagging, random forests, and boosting). Simulations based on two large cardiovascular datasets were used to evaluate the performance of MLTs in terms of MAPE, calibration, the C -statistic and Brier score, across six data-generating mechanisms (DGMs) and varying sample sizes. When the DGM and analysis model matched, boosting required a sample size 2–3 times larger than recommended; random forests and bagging did not achieve the target MAPE even with a 12-fold increase. For a neutral DGM that did not match any of the analysis models, logistic regression with only main effects and boosting resulted in target MAPE values with a 12-fold increase in the recommended sample size. For external validation, our simulations showed that sample size guidelines to achieve a target precision of the estimated C -statistic were suitable, and thus may be used to inform sample size calculations for MLTs.