Stroke Smart Healthcare Prediction

Description

This dataset is an integrated healthcare dataset derived from a multi-source Data Warehouse implemented using a Medallion architecture. The Gold layer represents the final analytical output and contains 143,960 patient records generated through the integration of three publicly available Kaggle datasets covering stroke, heart disease, and cardiovascular conditions. The dataset is designed to provide a unified and enriched representation of patient-level clinical and demographic information, enabling comprehensive healthcare analytics and research . The dataset is structured as a tabular dataset organized in a relational format suitable for analytical processing. It consolidates heterogeneous healthcare records into a single unified schema, ensuring consistent feature representation across all entries. Each record corresponds to an individual patient and captures multiple dimensions of health-related information, including demographic characteristics, clinical measurements, and lifestyle indicators. The integration process ensures that all data is harmonized and standardized, facilitating reliable comparative analysis and modeling. The final dataset consists of 25 features encompassing several categories. Demographic attributes include age, gender, marital status, work type, and residence type. Clinical measurements include glucose levels, body mass index (BMI), and blood pressure indicators. Cardiovascular-related features are aggregated to provide composite indicators such as average height, weight, systolic and diastolic pressure, cholesterol, and glucose levels. Additional heart-related metrics include resting blood pressure, maximum heart rate, and other derived indicators. A target variable is included to represent stroke occurrence. All features are encoded in numerical or categorical formats to support analytical workflows. The dataset was developed to support advanced healthcare analytics by integrating multiple data sources into a cohesive structure that reflects the complex relationships among cardiovascular and cerebrovascular risk factors. By combining diverse datasets into a unified analytical layer, it enables consistent analysis and supports the development of predictive and decision-support applications in healthcare contexts. The dataset is stored in tabular format and can be exported as CSV files, ensuring compatibility with common data analysis tools. All underlying data sources are publicly available and fully anonymized, containing no personally identifiable information. The dataset is intended strictly for research purposes and adheres to ethical standards for data usage in healthcare analytics.

Steps to reproduce

The Gold Layer is produced through a structured multi-stage ETL process within a Medallion Data Warehouse architecture. The process begins with loading multiple heterogeneous healthcare datasets (e.g., stroke, heart disease, cardiovascular) from Kaggle into the Bronze Layer in their raw format without transformation, ensuring full preservation of original data . In the Silver Layer, data preprocessing and harmonization are performed. This includes cleaning inconsistent records, removing duplicates, and handling missing values through imputation to ensure completeness. Data types are standardized across all datasets (e.g., converting categorical labels and numeric measurements into consistent formats). Feature names are unified to eliminate schema inconsistencies between sources. Categorical variables such as gender, work type, and residence are normalized into consistent encoded formats. Clinical measurements (e.g., glucose, BMI, blood pressure) are validated and scaled where necessary. Additionally, derived features are computed, such as aggregated cardiovascular indicators (e.g., average blood pressure values, combined cholesterol/glucose indicators). After cleaning, data integration is performed by merging all datasets into a unified schema. This involves aligning common attributes across datasets and resolving structural differences. Records are combined into a single dataset through union operations while ensuring consistent feature representation. Conflicting or redundant attributes are resolved by selecting standardized definitions, and missing attributes across sources are filled through transformation or derivation logic. In the Gold Layer, the final dataset is constructed as a single denormalized analytical table. All processed records are consolidated into one unified structure containing patient-level clinical, demographic, and lifestyle features. The dataset is validated to ensure zero missing values and consistency across all records. Final feature selection is applied to retain only relevant attributes required for analysis, resulting in a clean, feature-rich dataset. Finally, the Gold Layer dataset is stored in a tabular format and made available for downstream analytical use. It can be exported as CSV or accessed directly from the data warehouse environment, ensuring compatibility with machine learning and data analysis tools.

Institutions

• New Mansoura University

  Dakahlia, Al Mansurah
• Mansoura University

  Dakahlia, Al Mansurah

Categories

Related Links

Licence