Understanding the Public Leaderboard Score

The public leaderboard score is calculated as the simple average of four normalised metrics. Each metric contributes equally (25%) to the final score.

Targets and Metrics

• Target_purchase_next_1w – AUC

• Target_qty_next_1w – MAE

• Target_purchase_next_2w – AUC

• Target_qty_next_2w – MAE

Normalisation Rules

• AUC is already bounded between 0 and 1 and is used directly.

• MAE is normalised using min–max scaling and then inverted so that lower error results in a higher score.

Normalisation Ranges

• AUC: xmin = 0, xmax = 1

• MAE (Target_qty_next_1w): xmin = 0, xmax = 59.5561260135

• MAE (Target_qty_next_2w): xmin = 0, xmax = 81.6314093155

Benchmark Scores (see LB)

Target 1 WAUC = 0.837729787

Target 2 WAUC = 0.828651274

Target 1 Qty MAE = 0.616150905

Target 2 Qty MAE = 1.748611334

Normalised Scores (No Rounding Applied)

Target 1 WAUC normalised = 0.837729787

Target 2 WAUC normalised = 0.828651274

Target 1 Qty MAE normalised = 1 − (0.616150905 / 59.5561260135) = 0.9896542816626398

Target 2 Qty MAE normalised = 1 − (1.748611334 / 81.6314093155) = 0.9785791847934202

Final Score Formula

Final score = 0.25 × (Target 1 WAUC normalised + Target 2 WAUC normalised + Target 1 MAE normalised + Target 2 MAE normalised)

Final Public Leaderboard Score 0.908653631864015