Data from: Comparison of classification models for adolescent’s emotional and behavioral problems

Tilyeubai, Akhyt1 ; Tsend, Javzmaa 1 ; Uugandavaa, Nyamdavaa1 ; Vaanchindorj, Bayarmaa2 ; Luvsantseren, Purevdolgor1 ; Luvsan-Ish, Ajnai1 ; Chilkhaasuren, Baasandorj1 ; Chuluunbaatar, Galbadrakh1 ; Enkhjargal, Enkh-Urel1

Research facility: Mongolian National University of Medical Sciences

Published Sep 15, 2026 on Dryad. https://doi.org/10.5061/dryad.4xgxd25sm

Data files

Sep 15, 2026 version files 193.31 KB

README.md

12.18 KB
SDQ1_reproduce_JAMIA_Open_English.Rmd

7.90 KB
sdq1.csv

173.23 KB

Abstract

Objective: Mental disorders are very common among children and adolescents around the world. In Mongolia, the international Strengths and Difficulties Questionnaire (SDQ) is used to detect mental disorders in adolescents. In this paper, different types of classification methods were compared to determine adolescent emotional and behavioral problems using the SDQ.

Materials and methods: Data was collected from students, teachers, and parents, and databases were created for each group. The student database was divided into 10 folds by cross-validation, and the models were developed using classification methods and evaluated using their performance measures. The results were mainly analyzed using the Bayes model.

Results: The students 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: Future work will examine the difficulties encountered in the creation of emotional and behavioral problems among the students involved in the study using a progressive classification method.

Overview

This repository contains the dataset and R Markdown reproducibility file used to compare machine-learning classification models for identifying adolescents' emotional and behavioral problems.

The deposited files are:

sdq1.csv — the survey dataset used as the input data.
SDQ1_reproduce_JAMIA_Open_English.Rmd — the R Markdown analysis and reproducibility document.

The R Markdown file reads the data directly from sdq1.csv. The two files should therefore be kept in the same working directory when reproducing the analysis.

Descriptions

sdq1.csv

sdq1.csv is a comma-separated values (CSV) file containing 3,764 observations and 23 variables used in the analysis.

The deposited CSV contains age, gender, the 20 SDQ difficulty-item variables used in the analysis, and the Total Difficulties Score (TS). The five prosocial items (S1, S4, S9, S17, and S20) are not included in the deposited CSV.

Variable	Description	Score
`Age`	Age of the adolescent.	—
`Gender`	Gender of the adolescent.	—
`S2`	Restless, overactive, cannot stay still for long.	0–2
`S3`	Often complains of headaches, stomachaches, or sickness.	0–2
`S5`	Often has temper tantrums or hot tempers.	0–2
`S6`	Rather solitary, tends to play alone.	0–2
`S7`	Generally obedient, usually does what adults request.	0–2
`S8`	Many worries, often seems worried.	0–2
`S10`	Constantly fidgeting or squirming.	0–2
`S11`	Has at least one good friend.	0–2
`S12`	Often fights with other children or bullies them.	0–2
`S13`	Often unhappy, depressed or tearful.	0–2
`S14`	Generally liked by other children.	0–2
`S15`	Easily distracted, concentration wanders.	0–2
`S16`	Nervous in new situations, easily loses confidence.	0–2
`S18`	Often lies or cheats.	0–2
`S19`	Picked on or bullied by other children.	0–2
`S21`	Thinks things out before acting.	0–2
`S22`	Steals from home, school or elsewhere.	0–2
`S23`	Gets along better with adults than with other children.	0–2
`S24`	Many fears, easily scared.	0–2
`S25`	Good attention span, sees work through to the end.	0–2
`TS`	Total Difficulties Score used as the target variable in the classification analysis.	Classification target

In the deposited CSV, the SDQ item responses are represented numerically using scores from 0 to 2.

SDQ item groups and derived subscale scores

The SDQ subscale scores (B1–B4) are derived scores and are not stored as separate columns in sdq1.csv. They are defined from the corresponding SDQ item variables.

SDQ subscale	Variables	Description	Score
Emotional Symptoms (B1)	`S3, S8, S13, S16, S24`	Measures emotional symptoms such as worrying, nervousness, fears, unhappiness, and physical symptoms such as headaches or stomachaches.	Sum of the five scored item responses
Conduct Problems (B2)	`S5, S7, S12, S18, S22`	Measures conduct-related difficulties such as temper, obedience, fighting, lying or cheating, and taking things that do not belong to the adolescent.	Sum of the five scored item responses
Hyperactivity / Inattention (B3)	`S2, S10, S15, S21, S25`	Measures hyperactivity and attention-related difficulties, including restlessness, distractibility, concentration, thinking before acting, and ability to finish tasks.	Sum of the five scored item responses
Peer Relationship Problems (B4)	`S6, S11, S14, S19, S23`	Measures difficulties in relationships with peers, including preference for being alone, friendships, peer acceptance, bullying, and relationships with same-age peers.	Sum of the five scored item responses
Total Difficulties Score (TS)	`B1, B2, B3, B4`	Overall measure of emotional symptoms, conduct problems, hyperactivity/inattention, and peer relationship problems.	`TS = B1 + B2 + B3 + B4`

The standard SDQ also includes a Prosocial Behaviour subscale (B5) based on S1, S4, S9, S17, S20. These five prosocial item variables are not included in the deposited CSV and therefore are not available for deriving B5 from this dataset. Prosocial Behaviour is not included in the Total Difficulties Score. The deposited TS variable is used as the classification target in the reproducibility analysis.

Target variable

The TS variable contains three classification categories:

TS value	Classification
`0`	Normal
`1`	Borderline
`2`	Abnormal

For the binary classification analysis, only TS = 1 (Borderline) and TS = 2 (Abnormal) are selected. These are converted to the factor levels Borderline and Abnormal, respectively. The original TS variable is then removed before model training.

SDQ1_reproduce_JAMIA_Open.Rmd

SDQ1_reproduce_JAMIA_Open_English.Rmd is the R Markdown analysis and reproducibility document for the classification experiments.

The document performs the following steps:

Loads the required R packages.
Reads sdq1.csv.
Examines the dataset dimensions, variable names, and structure.
Examines the distribution of the target variable TS.
Selects the Borderline (TS = 1) and Abnormal (TS = 2) classes for binary classification.
Converts the selected target values into the factor levels Borderline and Abnormal.
Removes the original TS variable from the predictor dataset.
Checks predictor variables, summary statistics, and missing values.
Creates a stratified 10-fold cross-validation partition using set.seed(123).
Reuses the same fold partitions for all three classification models.
Trains and evaluates the C5.0 classifier.
Trains and evaluates the Naive Bayes classifier.
Trains and evaluates the RIPPER (JRip) classifier.
Produces confusion matrices for the three models.
Calculates Accuracy, Sensitivity, Specificity, and Kappa.
Compares the reproduced performance values with the values reported in the article.
Identifies the model with the highest reproduced Accuracy.

The R Markdown document uses the same cross-validation folds for all three models so that their performance is evaluated on comparable test partitions.

Relationship between the files

The relationship between the deposited files is:

sdq1.csv
   │
   │  read_csv("sdq1.csv")
   ▼
SDQ1_reproduce_JAMIA_Open.Rmd
   │
   ├── Data inspection
   ├── Target variable distribution
   ├── Borderline + Abnormal class selection
   ├── Predictor checking
   ├── 10-fold stratified cross-validation
   ├── C5.0
   ├── Naive Bayes
   ├── RIPPER (JRip)
   ├── Confusion matrices
   ├── Accuracy / Sensitivity / Specificity / Kappa
   └── Comparison with published results

The sdq1.csv file is the input dataset, while SDQ1_reproduce_JAMIA_Open.Rmd contains the analysis and reproducibility workflow.

Key information about the analysis

The analysis uses three classification algorithms:

Model	Description in the analysis
C5.0	Classification model trained using the `C5.0` R package.
Naive Bayes	Probabilistic classification model implemented using `e1071`.
RIPPER (JRip)	Rule-based classification model implemented through `RWeka`.

All three models use the same stratified 10-fold cross-validation partitions. The positive class for the reported confusion-matrix metrics is Abnormal.

Code / Software

R is required to run SDQ1_reproduce_JAMIA_Open.Rmd.

The following R packages are required:

readr
caret
C50
e1071
RWeka
dplyr
knitr

The R Markdown document checks for missing packages and installs them when necessary.

The RIPPER/JRip model requires a compatible Java/JDK installation together with the RWeka package.

Reproducing the analysis

Install R and a suitable R Markdown environment.
Place sdq1.csv and SDQ1_reproduce_JAMIA_Open.Rmd in the same folder.
Open SDQ1_reproduce_JAMIA_Open.Rmd.
Ensure the required R packages are available.
Ensure a compatible Java/JDK installation is available for RWeka.
Knit/render the R Markdown document to HTML.
The document will read sdq1.csv, perform the binary classification analysis, and generate the reported performance metrics.

The analysis uses set.seed(123) before creating the stratified 10-fold cross-validation folds.

Reproducibility and published results

The R Markdown document includes the performance values reported in the associated article for reference. The newly calculated results are compared with these published values for Accuracy, Sensitivity, and Specificity.

The published reference values included in the R Markdown document are:

Model	Published Accuracy	Published Sensitivity	Published Specificity
C5.0	0.866	0.899	0.825
Naive Bayes	0.856	0.898	0.805
RIPPER	0.771	0.785	0.756

Small differences between reproduced and published results may occur because of differences in data-processing sequence, random seed, cross-validation partitions, software versions, algorithm versions, or model parameters.

Notes

The deposited sdq1.csv contains the 20 SDQ difficulty-item variables and the TS target variable. The subscale scores B1–B4 are derived scores and are documented here to explain the structure and scoring relationship of the deposited SDQ items; they are not separate columns in the deposited CSV. The five prosocial items (S1, S4, S9, S17, and S20) are not included in the deposited dataset.

Human subjects data

This dataset contains de-identified data collected from adolescent participants. Ethical approval for this study was obtained from the relevant institutional ethics committee, and informed consent was obtained from the participants and/or their legal guardians prior to data collection.

All personally identifiable information, including names, contact information, and any direct identifiers, has been removed from the dataset before deposition. The dataset contains only anonymized variables used for statistical analysis and machine learning modeling. The deposited data cannot be used to identify individual participants.