Deduplicate rows, impute missing values with 8+ numeric strategies (Mean, Median, KNN, MICE) and 4 categorical strategies, remove outliers via IQR / Z-score / Winsorize, fix skewness, and apply Yeo-Johnson power transform. Download a clean CSV or hand off directly to AutoML.

log1p, sqrt, Yeo-Johnson, percentile rank, outlier flag, and missing flag per numeric column. Plus binning, polynomial pairs, interaction terms, date extraction, and cyclical encoding (sin / cos). All transforms are fit on training data only — no leakage.

Four methods — Variance Threshold, Correlation Filter (drop >0.9 correlated), RFE (Random Forest), and SelectKBest (Mutual Info) — automatically prune irrelevant or redundant columns before training. Configurable top-K cutoff.

RF, XGBoost, LightGBM, and CatBoost compete via 5-fold cross-validation. The winner is selected automatically by F1 (classification) or MAE (regression). Optional Optuna tuning and SHAP explanation run on the winner.

TPE sampler runs up to 30 trials on the AutoML winner to find optimal hyperparameters. Tuning is optional and runs after model selection — not before — so it never inflates the competition score.

Every prediction comes with a SHAP bar chart showing which features drove the result and by how much. FE-derived columns are grouped back to their originals so you see source-feature influence, not transform noise.

Simple voting (VotingClassifier / VotingRegressor) or stacking with a meta-learner on top of the AutoML winners. Reduces variance and improves generalization over any single model.

Visual card canvas that orchestrates all 7 ML stages — Preprocessing, Feature Engineering, Feature Selection, AutoML, Optuna, SHAP, and Ensemble — into one sequential pipeline.

Watch your data transform in real time — chibi scientist characters process each ML stage with fluid animations. A cinematic walkthrough of the full pipeline.