MLSS 2015
Predictive Modeling Challenge

News

September 27, 2015

Winners' report on Online market purchase prediction competition and Solar energy production prediction have been published.

September 1, 2015

Winners are announced.

August 13, 2015

Quick start guide for Online market purchase prediction competition has been published.

July 26, 2015

Dataset for Solar energy production prediction has been updated to fix incorrect correspondence between datetime and weather information. Please download the new one if you have the old one. Sorry for the inconvenience.

July 24, 2015

Competitions are now launched!: Solar energy production prediction and Online market purchase prediction

July 18, 2015

Competition materials are to be prepared. When they are ready, we will make an announcement to the registered participants, and put links on this page.

July 9, 2015

MLSS 2015 Predictive Modeling Challenge page has been launched. Competitions will start from the middle of July. The registration site is now open.

Results

Solar energy production prediction

Name Affiliation Final score (Mean absolute error)
1st place winner Dmitry Kit ("Dmitry") University of Bath 0.1550
2nd place winner Artem Oboturov ("eraser") 0.1580
Runner-up Liyuan Xu ("ly9988") The University of Tokyo 0.1601

Online market purchase prediction

Name Affiliation Final score (AUC)
1st place winner Liyuan Xu ("ly9988") The University of Tokyo 0.9538
2nd place winner Luca Celotti ("Actla") Université de Genève 0.9424
Runner-up Viktoriia Oliinyk ("htoze") National Research University, Higher School of Economics 0.9324

Competition procedure

  • Each participant can submit up to 3 predictions a day.
  • The score given to each submission is an intermediate score, which is calculated using 50% of the test dataset.
  • The final scores are calculated using the remaining portion of the test data. Final ranks are determined according to the final scores.

Competition rules

  • You are not allowed to form a team; each participant must work alone.
  • You are not permitted to privately share source codes or any executable codes developed for this competition.
  • To receive the prize, winners must provide the source codes along with a report that describes their approach.
  • You can use the competition datasets only during the competition period and are not allowed to use them for other purposes.
  • You must not copy and/or distribute the competition dataset.
  • You must erase the competition datasets from your storage after the end of the competition.

Introduction

We are happy to announce the MLSS 2015 Predictive Modeling Challenge in collaboration with KYOCERA Communication Systems (KCCS), one of the MLSS gold sponsors.

In this challenge, we will run two competitions, one for prediction of energy production by a solar electric generating facility, and the other for purchase prediction in an EC store.

The challenge will run for one month starting from the middle of July. Only the MLSS 2015 attendants can participate in the challenge. The first-place winner for each tasks will get an Apple Watch, and the second-place winner will get a AR.Drone 2.0. Winners are expected to provide a report describing their approach and to provide codes to check the reproducibility of their results.

The challenge will be held using the “University of Big Data” competition platform for data science education. Participants must register using a Google account at the registration site. Participants should use the registration code “MLSS2015”+ their MLSS ID number.

Sign up with Google account

If you have any question about the challenge, please contact universityofbigdata [at] gmail [dot] com. We are looking forward to you taking this opportunity to join the challenge!