Inspiration
Prediction is difficult, especially about the future — as Niels Bohr is often famously quoted. Nowadays, a highly relevant challenge in this field emerges in grocery e-commerce. The sector has not only shown rapid expansion, recently experiencing five years of growth over five months (McKinsey, The State of Grocery Retail 2021 - North America, p. 5), but also poses great difficulties to platforms as regards production and logistics due to these quick developments.
We set about to tackle the problem by training a neural network based time-series sales forecasting model.
What it does
The project aims to predict sales based on recent data of a grocery e-commerce platform
How we built it
We first conducted an extensive data exploration analysis using different approaches on understanding different feature distribution characteristics. Afterwards, we built a first model that incorporated the NeuralProphet library which we used to train our time series forecasting model.
Challenges we ran into
Time constraints as well as being forced to being pragmatic and prioritising punctuality over perfectionism.
Accomplishments that we're proud of
That we learned a lot along the way, immersing ourselves in the field of sales forecasting, which had we previously didn't have experience in.
What we learned
Appreciating the aesthetics of pandas.
What's next for HackUPC Demand Forecasting
Incorporating future pricing decisions into the model to even further enhance our ability to bring value to our customers and other stakeholders.
Log in or sign up for Devpost to join the conversation.