Dataset for Analyzing Psychological Barriers to Smoking Cessation Using Machine Learning in Bangladesh
Description
This dataset accompanies the research "Analyzing Psychological Barriers to Smoking Cessation Using Machine Learning" and contains anonymized survey data collected from current and former smokers in Bangladesh. The study was conducted as part of a Bachelor of Science Final Year Design Project at the Department of Computer Science and Engineering, Daffodil International University. Data were collected using a structured 21-item questionnaire administered through both Google Forms and face-to-face interviews to increase participation from individuals with limited access to digital platforms. The questionnaire was independently reviewed by two medical experts using the Content Validity Index (CVI) prior to data collection. A total of 901 responses were collected. After removing records that were not eligible for predictive modelling, 673 model-ready responses were retained for analysis. The dataset contains demographic information, smoking history, previous quit attempts, and ten psychological variables measured on five-point Likert scales. The target variable indicates whether a participant relapsed within one month after a previous quit attempt. The dataset was developed to support research on smoking cessation, public health informatics, explainable artificial intelligence (XAI), and machine learning. It can be used for binary classification, feature selection, class imbalance research, predictive modelling, explainability analysis using SHAP, and benchmarking machine learning algorithms. All personally identifiable information has been removed before publication to protect participant privacy.
Files
Steps to reproduce
1. Study Design: Conduct a quantitative cross-sectional survey targeting current and former smokers in Bangladesh who have previously attempted to quit smoking. 2. Questionnaire Development: Develop a structured 21-item questionnaire covering demographic characteristics, smoking history, smoking cessation history, and psychological barriers to smoking cessation measured using five-point Likert scales. Validate the questionnaire through a formal Content Validity Index (CVI) review conducted by two independent medical experts. 3. Data Collection: Collect responses using a dual-channel approach consisting of Google Forms (online) and face-to-face paper-based interviews to improve participation and reduce digital access bias. Obtain informed consent from all participants before data collection. 4. Data Preparation: Combine all collected responses into a single dataset. Remove incomplete or ineligible responses, including participants who had never attempted to quit smoking. After data cleaning and preprocessing, retain the model-ready dataset for analysis. 5. Data Preprocessing: Encode categorical variables, transform ordinal responses into numerical representations where appropriate, standardize continuous features, and prepare the target variable indicating smoking relapse within one month after a quit attempt. 6. Feature Selection: Apply Chi-Square, ANOVA F-test, Mutual Information, and Random Forest Feature Importance methods to rank predictor variables and select the most informative features. 7. Machine Learning Analysis: Train and evaluate machine learning models including Logistic Regression, Random Forest, XGBoost, LightGBM, Support Vector Machine (RBF), Multi-Layer Perceptron, Gradient Boosting, Stacking Ensemble, GA-SVM, and GA-RF using stratified 5-fold cross-validation. Evaluate performance using Accuracy, Precision, Recall, F1-score, and ROC-AUC. 8. Explainability: Apply SHAP (SHapley Additive exPlanations) to the best-performing tree-based model to identify and interpret the most influential psychological factors associated with smoking relapse. 9. Software: The analysis was performed using Python with common data science libraries, including pandas, NumPy, scikit-learn, XGBoost, LightGBM, DEAP, SHAP, and Matplotlib.
Institutions
- Daffodil International UniversityDhaka Division, Dhaka