Drug Craving Classification Model with SHAP Interpretability for 629 Abstainers

Weiqi Zeng's study presents a machine learning model for classifying drug craving levels among 629 abstainers recruited from Compulsory Isolation Drug Rehabilitation Centers. The dataset likely contains 18 demographic and behavioral features used to train and evaluate seven algorithms, with Logistic Regression selected as the optimal model. The work was published on figshare in May 2026 under a CC-BY-4.0 license.

Use Cases

• Training classification models for drug craving levels based on demographic and behavioral features.
• Applying SHAP analysis to interpret feature importance in clinical prediction models.
• Benchmarking machine learning algorithms like Logistic Regression, XGBoost, and LightGBM on imbalanced clinical data.
• Studying the relationship between factors like drug use frequency, duration, and heroin use with craving severity.

Strengths

• The dataset is based on 629 valid samples from a study of 692 abstainers, with preprocessing steps like missing value imputation and outlier removal documented.
• Model performance is reported with specific metrics: 66.13% accuracy and a 0.85 micro-average AUC on an independent test set.
• The study employs SHAP for model interpretability, quantifying contributions of features like frequency and duration of drug use.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The data may reflect geographic and population bias inherent to the specific rehabilitation centers sampled.

Provenance

Source

Weiqi Zeng via figshare.

Collection Method

Data collected from 692 abstainers at Compulsory Isolation Drug Rehabilitation Centers, with craving assessed via a 34-item Drug Craving Scale.

Freshness

Last updated 2026-05-11 05:25:38; freshness should be verified.

Geography

Data likely originates from Compulsory Isolation Drug Rehabilitation Centers, though specific geographic location is not stated.