electricsheepafrica/heart-disease-cleveland

Heart Disease Dataset (Cleveland UCI)

A cleaned and enriched version of the classic UCI Cleveland Heart Disease Dataset, one of the most widely used benchmarks in medical machine learning research. This dataset contains clinical and diagnostic measurements from 302 patients used to predict the presence or absence of heart disease.

Dataset Summary

Property	Value
Source	UCI Machine Learning Repository — Cleveland Clinic Foundation
Kaggle Source	johnsmith88/heart-disease-dataset
Task	Binary Classification
Samples	302 (deduplicated from 1,025 raw rows)
Features	13 input features + 1 target
Target	target — 0 = No Disease, 1 = Heart Disease
Missing Values	None (anomalous encodings flagged)
License	Unknown (original UCI data is publicly available for research)

---

Background

The original Cleveland Heart Disease data was collected at the Cleveland Clinic Foundation and donated to the UCI ML Repository by David W. Aha in 1988. It has been used in hundreds of peer-reviewed publications on cardiovascular risk prediction, explainable AI, and clinical decision support.

The Kaggle version of this dataset contained 1,025 rows — the original 302 unique records inflated with 723 duplicate rows. This release removes all duplicates and restores the dataset to its canonical 302-sample form, with added human-readable label columns for each categorical feature.

---

Cleaning & Processing

The following steps were applied to produce heart_disease_cleaned.csv:

1. Deduplication — Removed 723 duplicate rows, restoring the 302-sample Cleveland dataset.
2. Anomaly flagging — Two rows with thal=0 and four rows with ca=4 were flagged (boolean columns thal_flag and ca_flag). These values fall outside the standard Cleveland encoding and may represent missing data in the original collection. They are retained with flags rather than silently dropped.
3. Label enrichment — Human-readable label columns were added for all categorical features (see schema below).
4. Column reordering — Columns are arranged semantically: demographics → cardiac symptoms → test results → target.

---

Schema

Numeric / Encoded Features

Column	Type	Description
age	int	Age of the patient in years
sex	int	Sex: 0 = Female, 1 = Male
cp	int	Chest pain type: 0 = Typical Angina, 1 = Atypical Angina, 2 = Non-Anginal Pain, 3 = Asymptomatic
trestbps	int	Resting blood pressure (mm Hg) on admission
chol	int	Serum cholesterol (mg/dl)
fbs	int	Fasting blood sugar > 120 mg/dl: 0 = No, 1 = Yes
restecg	int	Resting ECG results: 0 = Normal, 1 = ST-T Wave Abnormality, 2 = Left Ventricular Hypertrophy
thalach	int	Maximum heart rate achieved (bpm)
exang	int	Exercise-induced angina: 0 = No, 1 = Yes
oldpeak	float	ST depression induced by exercise relative to rest
slope	int	Slope of peak exercise ST segment: 0 = Upsloping, 1 = Flat, 2 = Downsloping
ca	int	Number of major vessels colored by fluoroscopy (0–3); value 4 flagged as anomalous
thal	int	Thalassemia: 1 = Normal, 2 = Fixed Defect, 3 = Reversible Defect; value 0 flagged as anomalous
target	int	Diagnosis of heart disease: 0 = No Disease, 1 = Heart Disease

Added Label Columns

Column	Description
sex_label	Human-readable sex label
cp_label	Human-readable chest pain type
fbs_label	Human-readable fasting blood sugar
restecg_label	Human-readable resting ECG result
exang_label	Human-readable exercise-induced angina
slope_label	Human-readable ST slope
thal_label	Human-readable thalassemia result
target_label	Human-readable diagnosis
thal_flag	True if thal=0 (encoding anomaly)
ca_flag	True if ca=4 (encoding anomaly)

---

Class Distribution

Class	Count	Percentage
Heart Disease (1)	164	54.3%
No Disease (0)	138	45.7%

The dataset is approximately balanced, making it suitable for standard binary classification without aggressive resampling.

---

Patient Demographics

Feature	Value
Age range	29 – 77 years
Mean age	54.4 years
Median age	55.5 years
Male patients	206 (68.2%)
Female patients	96 (31.8%)

---

Usage

from datasets import load_dataset

ds = load_dataset("electricsheepafrica/heart-disease-cleveland")
df = ds["train"].to_pandas()

print(df["target_label"].value_counts())

Or with pandas directly:

import pandas as pd

df = pd.read_csv("heart_disease_cleaned.csv")

# Features and target
X = df[["age","sex","cp","trestbps","chol","fbs","restecg",
        "thalach","exang","oldpeak","slope","ca","thal"]]
y = df["target"]

---

Example: Quick Baseline

from datasets import load_dataset
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score
import pandas as pd

ds = load_dataset("electricsheepafrica/heart-disease-cleveland", split="train")
df = ds.to_pandas()

feature_cols = ["age","sex","cp","trestbps","chol","fbs","restecg",
                "thalach","exang","oldpeak","slope","ca","thal"]
X = df[feature_cols]
y = df["target"]

clf = RandomForestClassifier(n_estimators=100, random_state=42)
scores = cross_val_score(clf, X, y, cv=5, scoring="accuracy")
print(f"CV Accuracy: {scores.mean():.3f} ± {scores.std():.3f}")

---

Known Issues & Limitations

• Encoding anomalies: 2 rows have thal=0 and 4 rows have ca=4, which are outside the standard Cleveland encoding range. These are flagged with thal_flag and ca_flag boolean columns.
• Small sample size: 302 samples may limit generalization. Consider augmentation or transfer learning for production clinical models.
• Single-site data: All patients were collected from the Cleveland Clinic Foundation. External validity to other populations should be evaluated carefully.
• No temporal data: This is a cross-sectional snapshot; no longitudinal follow-up is included.

---

Citation

If you use this dataset, please cite the original UCI source:

Detrano, R., Janosi, A., Steinbrunn, W., Pfisterer, M., Schmid, J., Sandhu, S., Guppy, K., Lee, S., & Froelicher, V. (1989).
International application of a new probability algorithm for the diagnosis of coronary artery disease.
American Journal of Cardiology, 64(5), 304-310.

Dua, D. and Graff, C. (2019). UCI Machine Learning Repository. Irvine, CA: University of California,
School of Information and Computer Science. http://archive.ics.uci.edu/ml

---

Collection

This dataset is part of the Electric Sheep Africa — Healthcare Collection, a curated set of medical and clinical datasets for research and education.

Explore the full collection: electricsheepafrica