Machine Learning contest IMDC

  • awards
  • general
03/04/2020

IMDC hosted an internal machine learning (ML) competition to predict beach erosion volumes caused by waves– an initiative of our colleague Gert Leyssen. Every one of the company was invited and 21 participants joined the competition. 

A training dataset was provided and the teams worked on the two main blocks of every ML: 

  1. Improvement of the predictor variables (with engineering knowledge of beach erosion, derive new relations, …) 
  2. The selection and training of the best ML regressor. Multiple techniques ranging from k-nearest neighbors till boosted tree regressors and neural networks were applied and trained. 

A website was hosted to upload and compare the results of the predictions of an independent validation dataset so every team could see their score and rank. Finally the winner of the competition was Jasper Claus and his team with a model based on the Extra trees regressor and they won a Raspberry Pi 4. The price for the best alternative approach went to Team BlackMamba (aka Bart Pannemans). 

 It was interesting to see how in a relative short time the combination of expert knowledge and an enthusiastic group could result in nice results.

Correlation analysis to investigate the relations between the predictors
Relations and histograms of the predictors
Prediction results of the winning model (divided in storm and normal conditions)