Current Season: Day 3

On the third day of the news season NielsenIQ at GfK - An NIQ Compan hosts an exciting machine learning challenge in the context of customer credits.

Thank you Jeff from GFK for designing such an innovative challenge, that seemed to be easy at the beginning, but revealed its hardness throughout the day.

More Info on the Problem

NIQ Waj´kanda needs your help. The villain “De Danos” hinders Vibranium Glide, the biggest retailer of hoverboards, to deliver us sales data, which will challenge the accuracy of our reporting. However, other retailers are not affected by this iniquity and continue to send their data. You have been selected to be part of a taskforce in charge of modelling Vibranium Glide’s monthly sales after they stopped their regular data deliveries. To help in this process you will have access to the full history of Vibranium Glide’s sales, alongside historical (for training) and still ongoing (for inference) data deliveries from its four main competitors: Panther Hover, Wakanda Wheels, Shuri’s Skates and HoverTech Wakanda. These data report for each retailer the monthly sales of all known models of hoverboard. The final delivery will be simulation of Vibranium Glide’s hoverboard sales at various horizons in a scheme that mimics real-time and prevents look ahead bias:

• Month t, i.e. the first month after delivery stopped. This prediction will have to be submitted by 3pm.
• Months t+1 and t+2, i.e. the two following months. You will get the necessary data of the competitors and submission template after handing in predictions for Month t. This prediction will have to be submitted by 4pm.
• Months t+3, t+4 and t+5, i.e. the three following months. You will get the necessary data of the competitors and submission template after handing in predictions for Months t+1 and t+2. This prediction will have to be submitted by 5pm. This time we will not provide explicit feedback on predictions but you can choose to use the last months of training data as validation sets as demoed in the provided evaluation notebook (see below for available data and resources). From a methodological point of view, this exercise is a hierarchical time series modelling and prediction problem (e.g. see this article on Medium https://medium.com/data-science/a-simple-approach-to-hierarchical-time-series- forecasting-with-machine-learning-2e180d83966c). Therefore, simulations will be evaluated in terms of sales units per model, brand and overall.

Impressions

Results

We’re thrilled to announce that NKD Group has won the third matchday of this season, hosted by NielsenIQ at GfK - An NIQ Company. The task was out of the comfort zone of most participants and pushed each team to explore new approaches in time-series analyses and forecasting.

A huge congratulations to the team from NKD Group for their remarkable performance! 🎉