Tabular Datasets Intermediate

Structured, tabular datasets remain the most common type of data in real-world machine learning applications. From the UCI Repository classics to Kaggle competition favorites and government open data.

UCI Machine Learning Repository

The UCI ML Repository hosts 600+ datasets and has been a cornerstone of ML research since 1987. Top datasets:

Kaggle Popular Datasets

House Prices

1,460 homes, 79 features. The "Ames Housing" dataset. Predict sale price from features like lot area, year built, number of bedrooms, garage type. The classic regression competition.

import pandas as pd

# Download via Kaggle API
# kaggle competitions download -c house-prices-advanced-regression-techniques
train = pd.read_csv("train.csv")
print(train.shape)  # (1460, 81)

Credit Card Fraud

284,807 transactions, 492 fraudulent (0.17%). Highly imbalanced binary classification. Features are PCA-transformed for privacy. A standard benchmark for anomaly detection and imbalanced learning.

Customer Churn

7,043 customers. Predict whether a telecom customer will leave. Features include tenure, monthly charges, contract type, and services subscribed.

Spotify Tracks

114,000+ tracks. Audio features (danceability, energy, tempo, valence) for Spotify songs. Used for music recommendation, genre classification, and popularity prediction.

Netflix Shows

8,800+ titles. Netflix catalog including type (movie/show), director, cast, country, release year, rating, duration. Used for content analysis and recommendation systems.

Airbnb Listings

Varies by city (10K-50K+ listings). Inside Airbnb provides listing data including price, location, reviews, amenities, and availability. Used for price prediction and market analysis.

Government Open Data

Financial Datasets

• Yahoo Finance: Historical stock prices via yfinance Python library
• FRED: Federal Reserve Economic Data (interest rates, GDP, unemployment)
• Quandl: Financial and economic time series data

Healthcare Datasets

• MIMIC-III/IV: Critical care database with 40,000+ ICU stays. Requires credentialing through PhysioNet.
• NIH Chest X-rays: 112,120 frontal-view X-ray images with disease labels
• PhysioNet: Various physiological signal datasets (ECG, EEG, etc.)