{
  "split": "Train: booking before July 2016 and outcome known before July 2016. Validate: July\u2013September 2016 bookings with outcomes known before January 2017. Test: January\u2013April 2017 bookings, all recorded outcomes available in the published August 2017 extract.",
  "train_rows": 54296,
  "test_rows": 19986,
  "selection": "Logistic regression selected by validation log loss. Fixed validation 90th-percentile score threshold. No test tuning.",
  "features": [
    "LeadTime",
    "ArrivalMonth",
    "ArrivalWeekday",
    "StaysInWeekendNights",
    "StaysInWeekNights",
    "Adults",
    "Children",
    "Babies",
    "Hotel",
    "Meal",
    "MarketSegment",
    "DistributionChannel",
    "ReservedRoomType"
  ],
  "target": "Recorded cancellation indicator",
  "validation": [
    {
      "model": "Logistic regression",
      "validation_log_loss": 0.5541416627667153,
      "validation_brier": 0.18603475752436704
    },
    {
      "model": "Gradient boosting 15",
      "validation_log_loss": 0.5672474071353542,
      "validation_brier": 0.19359081343554116
    },
    {
      "model": "Gradient boosting 31",
      "validation_log_loss": 0.5756277358947838,
      "validation_brier": 0.19694975671202145
    }
  ],
  "feature_availability": "Conservative snapshot predictors; arrival, party, channel and reserved room may still have changed after booking. Event-versioned source data are required before deployment.",
  "excluded_features": [
    "ReservationStatus",
    "ReservationStatusDate",
    "AssignedRoomType",
    "BookingChanges",
    "DepositType",
    "ADR",
    "Agent",
    "Company",
    "Country",
    "DaysInWaitingList",
    "TotalOfSpecialRequests",
    "RequiredCarParkingSpaces",
    "CustomerType",
    "PreviousCancellations",
    "PreviousBookingsNotCanceled"
  ],
  "limits": "No customer identifier for entity-separated evaluation. Repeated source records retained because groups can contain legitimately identical bookings. Cohort composition changes. No treatment or profit data."
}