Journal of Psychosociological Research in Family and Culture

Objective: The present study aimed to predict fertility intentions based on cultural norms, economic security, and relationship satisfaction using advanced machine learning techniques.

Methods and Materials: This study employed a cross-sectional, predictive-correlational design conducted on 512 adults of reproductive age in South Africa selected through stratified random sampling. Data were collected using standardized instruments including the Cultural Values Scale, Economic Stability Index, Dyadic Adjustment Scale, and a Fertility Intention Scale, all of which demonstrated acceptable validity and reliability in previous studies. Data were analyzed using IBM SPSS-27 for descriptive and correlational analyses, followed by machine learning modeling in Python using Random Forest, Support Vector Machine, Gradient Boosting, and Artificial Neural Network algorithms. Data preprocessing included normalization, missing data imputation, and categorical encoding. Model performance was evaluated using 10-fold cross-validation and metrics including accuracy, precision, recall, F1-score, and area under the ROC curve.

Findings: The results indicated significant positive relationships between cultural norms, economic security, relationship satisfaction, and fertility intentions (p < .01). Among predictors, economic security demonstrated the strongest predictive power, followed by relationship satisfaction and cultural norms. Machine learning results showed that the Artificial Neural Network achieved the highest performance (accuracy = 0.868, AUC = 0.926), followed by Gradient Boosting (accuracy = 0.856, AUC = 0.914) and Random Forest (accuracy = 0.842, AUC = 0.901), while Support Vector Machine showed comparatively lower performance (accuracy = 0.801, AUC = 0.862). Feature importance analysis confirmed the dominant role of economic security across all models.

Conclusion: The superior performance of machine learning models demonstrates their effectiveness in capturing complex, nonlinear relationships among predictors, offering a powerful approach for understanding and predicting reproductive decision-making.

Objective: The present study aimed to identify and model the cultural determinants of help-seeking behavior using machine learning techniques, focusing on stigma internalization, norm salience, and self-construal.

Methods and Materials: This cross-sectional predictive study was conducted among 512 adults residing in Canada, selected through a stratified sampling approach to ensure demographic diversity. Data were collected using standardized instruments, including the General Help-Seeking Questionnaire (GHSQ), the Internalized Stigma of Mental Illness Scale (ISMI), a validated Social Norms Salience Scale, and the Self-Construal Scale (SCS). Data preprocessing involved normalization, handling missing values via multiple imputation, and encoding categorical variables. Statistical analyses were performed using SPSS-27 for descriptive and correlational assessments. Machine learning models, including Logistic Regression, Support Vector Machine, Random Forest, and Gradient Boosting, were implemented in Python using Scikit-learn and XGBoost. Model performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics, with k-fold cross-validation ensuring robustness. Feature importance was examined using SHAP (SHapley Additive Explanations) to determine the relative contribution of predictors.

Findings: The results indicated that stigma internalization was the strongest negative predictor of help-seeking behavior, while norm salience and self-construal showed significant positive predictive effects. Gradient Boosting demonstrated the highest performance (AUC = 0.92), outperforming other models. Correlation analysis revealed significant negative associations between stigma internalization and help-seeking (p < .01), and positive associations for norm salience and self-construal (p < .01). SHAP analysis confirmed the hierarchical importance of predictors, with stigma internalization exerting the greatest influence, followed by norm salience and self-construal, and revealed nonlinear interaction effects among variables.

Conclusion: The findings highlight the central role of culturally embedded psychological factors in shaping help-seeking behavior and demonstrate the effectiveness of machine learning approaches.

Objective: The present study aimed to identify and model latent cultural value system profiles using machine learning techniques based on collectivism, power distance, and uncertainty avoidance.

Methods and Materials: This study employed a cross-sectional, descriptive–analytical design with a predictive modeling approach. The sample consisted of 512 adult participants from Canada selected through stratified random sampling to ensure demographic and cultural diversity. Data were collected using standardized instruments measuring collectivism, power distance, and uncertainty avoidance, all of which demonstrated established validity and reliability in prior research. After data preprocessing, including normalization and handling of missing values, both supervised and unsupervised machine learning techniques were applied. Classification models included support vector machines, random forest, gradient boosting, and logistic regression, while k-means clustering was used to identify latent cultural profiles. Model evaluation was conducted using stratified k-fold cross-validation, with performance metrics including accuracy, precision, recall, F1-score, and AUC-ROC. Feature importance and interpretability were assessed using SHAP analysis.

Findings: The results indicated significant positive associations among collectivism, power distance, and uncertainty avoidance (p < 0.01). Among classification models, gradient boosting demonstrated the highest predictive performance (AUC-ROC = 0.927), followed by random forest (AUC-ROC = 0.912), indicating strong model discrimination. Logistic regression showed comparatively lower performance, suggesting the presence of nonlinear relationships among variables. Clustering analysis identified three distinct cultural profiles characterized by low, moderate, and high levels of the examined dimensions. Feature importance analysis revealed that power distance was the strongest predictor of cultural profile classification, followed by uncertainty avoidance and collectivism.

Conclusion: The study underscores the importance of power distance and uncertainty avoidance in shaping cultural profiles and supports the utility of advanced computational methods in cultural research.

Objective: The present study aimed to model and predict risk-taking behavior by examining the interactive effects of sensation seeking, norm deviance, and peer influence within a cultural framework using machine learning techniques.

Methods and Materials: This study employed a cross-sectional, predictive-correlational design with a sample of 462 young adults from Greece selected through stratified sampling. Data were collected using standardized self-report instruments assessing risk-taking behavior, sensation seeking, norm deviance, and peer influence. After data preprocessing, including normalization and missing data imputation, both statistical and machine learning analyses were conducted. Pearson correlations were used to examine associations among variables, followed by the implementation of multiple supervised machine learning models, including Random Forest, Support Vector Machine, Gradient Boosting, and Artificial Neural Networks. The dataset was divided into training and testing subsets using an 80/20 split, and 10-fold cross-validation was applied to enhance model generalizability. Model performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics, while SHAP analysis was used to determine feature importance and interpret model predictions.

Findings: Inferential analyses indicated significant positive relationships between risk-taking behavior and sensation seeking (r = 0.54, p < 0.01), peer influence (r = 0.52, p < 0.01), and norm deviance (r = 0.49, p < 0.01). Machine learning results revealed that the Gradient Boosting model demonstrated the highest predictive performance (accuracy = 0.88, AUC-ROC = 0.93), followed by Random Forest and Neural Network models. Feature importance analysis using SHAP values showed that sensation seeking was the strongest predictor (mean SHAP = 0.37), followed by peer influence (0.31) and norm deviance (0.28), indicating that both individual and social factors significantly contribute to the prediction of risk-taking behavior.

Conclusion: The superior performance of ensemble machine learning models highlights the importance of capturing nonlinear and complex relationships among predictors.

Objective: The present study aimed to model and predict family decision-making effectiveness using machine learning techniques by examining the roles of shared mental models, power dynamics, and negotiation styles.

Methods and Materials: This descriptive–correlational study with a predictive modeling approach was conducted on 412 married adults in Thailand selected through multistage cluster sampling. Data were collected using standardized instruments assessing shared mental models, family power dynamics, and negotiation styles, all of which demonstrated acceptable validity and reliability. Data analysis was performed using a hybrid approach combining statistical analysis in IBM SPSS Statistics (version 27) and machine learning modeling in Python with scikit-learn and TensorFlow. Predictive models including Random Forest, Support Vector Machine, Gradient Boosting, and Multilayer Perceptron were trained and evaluated using cross-validation, with performance assessed through accuracy, precision, recall, F1-score, and AUC-ROC metrics.

Findings: The results indicated that shared mental models had a significant positive effect on family decision-making effectiveness, while power dynamics showed a significant negative effect. Integrative negotiation style significantly and positively predicted decision-making effectiveness, whereas dominating style had a significant negative association. Machine learning analysis revealed that the Gradient Boosting model achieved the highest predictive performance (accuracy = 0.89, AUC = 0.92), outperforming other models. Feature importance analysis demonstrated that shared mental models were the strongest predictor, followed by integrative negotiation style and power dynamics, confirming the relative contribution of cognitive, behavioral, and structural variables in predicting decision outcomes.

Conclusion: The findings highlight the central role of cognitive alignment and collaborative negotiation in enhancing family decision-making effectiveness, while unequal power structures undermine optimal outcomes. The integration of machine learning approaches provides a robust and nuanced framework for modeling complex family processes, offering both theoretical advancement and practical implications for improving relational functioning.

Objective: The present study aimed to predict adolescent risk behavior by examining the combined effects of family monitoring, sensation seeking, and peer deviance using machine learning analytical approaches.

Methods and Materials: This cross-sectional predictive study was conducted among 512 adolescents aged 14 to 18 years recruited from secondary schools in Tunisia using multistage cluster random sampling. Data were collected using standardized self-report instruments, including the Youth Risk Behavior Surveillance Questionnaire, the Parental Monitoring Scale, the Brief Sensation Seeking Scale, and the Peer Delinquency Scale, all of which demonstrated acceptable reliability and validity. Statistical analyses were performed using SPSS-27 to compute descriptive indices and Pearson correlations. Machine learning models, including random forest, support vector machine, gradient boosting, and multilayer perceptron neural networks, were implemented in Python. Data preprocessing included normalization and feature scaling, and model performance was evaluated using k-fold cross-validation and metrics such as accuracy, precision, recall, F1-score, and area under the ROC curve.

Findings: The results revealed that adolescent risk behavior was significantly negatively associated with family monitoring and positively associated with sensation seeking and peer deviance. Among the predictive models, the neural network demonstrated the highest performance, achieving the greatest accuracy and discriminative power. Feature importance analyses consistently identified peer deviance as the strongest predictor of risk behavior, followed by sensation seeking, while family monitoring showed a weaker but still significant contribution. The models indicated that nonlinear interactions among predictors significantly improved prediction accuracy compared to linear approaches.

Conclusion: The findings highlight the critical role of peer deviance and sensation seeking in shaping adolescent risk behavior, while confirming the protective function of family monitoring. The superior performance of machine learning models underscores their utility in capturing complex behavioral patterns and enhancing predictive precision.