🧠 How It Works

This app runs a full data profiling and quality audit using statistical methods — all locally in your browser session. No code, no setup, just results.

• 📋 Comprehensive Column Classification
  Columns are automatically categorised into numeric, boolean, categorical, datetime, or timeseries types. Timeseries checks include monotonicity and time span detection.
• 📊 Descriptive Statistics
  For every numeric column, we compute mean, standard deviation, min, max, variance, skewness, and kurtosis. Outliers are detected using Z-scores with thresholds at 3σ, 4σ, and 5σ.
• 📈 Distribution Diagnostics
  Skewed columns are flagged and classified (moderate or severe). You'll get transformation suggestions like log, Box-Cox, and Yeo-Johnson — with histograms for visual inspection.
• 🚨 Outlier Detection & Visualisation
  Outliers are split by severity level and shown with plots. This helps identify influential points or data errors before they affect your models.
• 🧪 Data Quality Audit
  The app automatically checks for:
  • ✔️ Missing data (0–29% or 30%+ severity)
  • ✔️ Fully null columns
  • ✔️ Duplicate rows
  • ✔️ Constant and low-variance columns (including categorical and boolean)
  • ✔️ High and medium cardinality features
  • ✔️ Imbalanced boolean features (over 70% one class)
• 🔗 Correlation Analysis
  Computes Pearson correlation for numeric features and Cramér’s V for categoricals. Visualises numeric correlation heatmaps and flags highly collinear pairs.
• 🧮 Multicollinearity Detection (VIF)
  Applies preprocessing (null filtering, constant drop, imputation), then calculates Variance Inflation Factors. Warns about numeric features with VIF > 5 or 10.
• 💡 Actionable Insights
  Recommendations are shown for each issue — complete with example code, severity badges, and justifications so you can clean data efficiently.

🔐 All data stays in memory — nothing is stored or shared. This is a fully stateless, secure analysis workflow.