Background: Likert-scale questionnaire data is widely used to evaluate satisfaction; however, in practical applications, both high classification accuracy and model interpretability are essential requirements. Rule-based machine learning methods provide the opportunity to satisfy both criteria simultaneously. Despite the rapid growth of research using machine learning for classification and prediction in recent years, there is a scarcity of studies employing the C5.0 classification algorithm for Likert-scale satisfaction data, which serves as the rationale for this study. Aim: To implement the C5.0 classification algorithm using a 10-fold cross-validation method to predict two levels of satisfaction (High/Medium) and to evaluate the model’s performance and interpretability. Materials and Methods: he study was conducted at the Department of Physics and Informatics, School of Bio-Medicine, MNUMS, utilizing satisfaction survey data from graduates majoring in Health Information Technology. Data consisting of four Likert-scale variables (V1–V4) were used. The C5.0 model was trained using a parametric grid search and evaluated within a 10-fold cross-validation framework using accuracy and a confusion matrix. Furthermore, IF–THEN rules were extracted from the final rule-based model, and the decision logic was visualized in the form of a flowchart. Results: The results indicated that the rule-based C5.0 model was selected as the optimal model, demonstrating stable performance throughout the cross-validation process. The confusion matrix results showed high precision in identifying the “High” satisfaction class. Among the generated rules, it became evident that variables V4 and V2 have the most significant influence on determining satisfaction levels. Conclusion Utilizing the C5.0 algorithm within a cross-validation framework enables effective and intelligible classification of satisfaction based on Likert data. It was confirmed that the rule-based structure offers distinct advantages for model interpretation and decision-support applications.

Background: Globally, an estimated 13% of adolescents aged 10–19 are affected by mental disorders. As of 2020, the number of children under the age of 19 in Mongolia was reported to be 1,289,587, reflecting a 0.9% increase compared to 2015. A 2013 study on adolescents found that 60.5% were categorized as mentally healthy, 30.5% exhibited emotional and behavioral difficulties, and 9% were diagnosed with a mental disorder. A study conducted in Govi-Altai Province between 2018 and 2019 reported that 58.6% of adolescents were classified as healthy, 36.1% experienced psychological difficulties, and 5.2% were diagnosed with a mental disorder. Internationally, adolescent mental health has been widely assessed using the Strengths and Difficulties Questionnaire (SDQ), with an increasing number of studies employing artificial intelligence-based predictive models. However, in Mongolia, research utilizing artificial intelligence and machine learning for analyzing large-scale mental health data remains limited. This gap underscores the need for the present study. Aim: Machine learning models were compared to determine adolescent emotional and behavioral problems using the SDQ. Materials and Methods: Data was collected from teenagers, teachers, and parents in Govi-Altai Province, and the databases were created for each group. The teenager database was divided into 10 folds by cross-validation, and the models were developed using machine learning methods and evaluated using their performance measures. The results were mainly analyzed using the Bayes model. Results: The teenagers have emotional and behavioral problems due to emotional and peer interactions, but they are at risk of developing disorders due to hyperactivity and behavioral changes. Conclusion Comparing the model performance results with previous studies, Bayesian model accuracy decreased by 0.03, sensitivity decreased by 0.08, and specificity increased by 0.01. Also, the difference between the performance evaluation metrics of the C50 and Bayesian models is very small, between 0.01 and 0.02. This shows that the performance of the Bayesian method is good when the number of attributes in the database increases. Compared to the results of the knowledge generated by the research, the participants are more likely to develop emotional and behavioral disorders due to their peer relationship indicators, such as other children generally not liking them, getting on better with adults, and due to emotional symptoms such as being unhappy and depressed.