Hypothyroidism Prediction Dataset with 3,772 Clinical Records

3,772 clinical records with 30 features form a public dataset used to develop a deep learning model for hypothyroidism prediction. The dataset was used by author Divya Kesavulu to train the FusionNet-CXG model, with results published on figshare in April 2026. Performance metrics for the model, including a mean accuracy of 0.9394, were obtained through 50 fold-level evaluations using SMOTE-NC and cross-validation.

Use Cases

• Training diagnostic classifiers for hypothyroidism based on 30 clinical features.
• Benchmarking hybrid deep learning architectures like CNN-LSTM-GRU models on medical tabular data.
• Conducting feature importance analysis for clinical predictions using methods like SHAP.

Strengths

• Contains 3,772 clinical records, providing a substantive sample for model development.
• Includes 30 features per record, suggesting a multi-faceted view of patient data.
• Model performance was evaluated using a rigorous protocol of 10 repetitions of 5-fold stratified cross-validation.

Limitations

• Row count for the specific dataset file is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect source bias inherent to the original public dataset; external validation was noted as future work.

Provenance

Source

A public dataset, source unspecified.

Collection Method

Originally gathered for clinical research; processed with SMOTE-NC for class imbalance.

Freshness

Last updated 2026-04-21 17:49:05