/dextra-mindef-2015

My solution for Dextra Data Science Challenge #44 (Singapore Ministry of Defense) https://challenges.dextra.sg/challenge/44

Primary LanguagePython

Ministry of Defence Data Analytics Challenge

img

img

Note: A few people asked me for the challenge's data source. Unfortunately, I am not authorized to publicly release it - if you need it, please do send the request to either Mindef or Dextra.sg instead of sending it to me.

Challenge URL: http://www.dextra.sg/ministry-of-defence-data-analytics-challenge/

Quick analysis with Tableau Public [Removed due to non-disclosure agreement]

Libraries used: Scikit-Learn, Pandas, XGBoost, Mathplotlib

Scores:

Public Leader Board (5th/158 Participants)

  • 0.0169515: best Single XGBoost model
  • 0.0168939: blending multiple XGBoost models with different Features Set.

Private Leader Board (1st/158 Participants)

  • 0.0141351

Submission History (only the best one):

Only Native XGBoost was recorded since it just dominated everything.

  1. Public Leader Board 0.0171364

    $ python classify-xgb-native.py # 990r depth6 0.0155765992602 0.019516592639 0.00988590074655 0.0141124661651 0.014303086534 Mean: 0.014678929069 (Local Score)

  2. Public Leader Board 0.0172253

    $ python classify-xgb-native.py # 180r 0.0157726389016 0.0201645979107 0.0095532522597 0.013888759618 0.0139117869773 Mean: 0.0146582071335 (Local Score)

  3. Public Leader Board 0.0171475

    $ python classify-xgb-native.py #added age_gender, rm a bunch of features 0.015551655811 0.019148557532 0.00965389534226 0.0139233429833 0.0139280448029 Mean: 0.0144410992943 (Local Score)

  4. Public Leader Board 0.0171112

    $ python classify-xgb-native.py # promo - gender 0.0155083548415 0.0189263516813 0.00951782504063 0.0140093232169 0.014178032663 Mean: 0.0144279774887 (Local Score)

  5. Public Leader Board 0.0170703

    $ python classify-xgb-native.py # cap salary 101% 0.0153414063482 0.0189991328711 0.00959486331913 0.0139794582592 0.0140253377611 Mean: 0.0143880397117 (Local Score)

  6. Public Leader Board 0.0170369

    $ python classify-xgb-native.py # INJURY TYPE as String 0.0153022751895 0.0189944794534 0.00957494483944 0.0139220394066 0.014069437855 Mean: 0.0143726353488 (Local Score)

  7. Public Leader Board 0.0169515

    $ python classify-xgb-native.py # better minchildage # treat as str 0.0152455036731 0.0189285563506 0.00961418416464 0.0139189502782 0.0139664367926 Mean: 0.0143347262518 (Local Score)