Machine Learning as it Relates to ERP
The purpose of this page is to assemble knowledge and resources to help ERP Academy members maximize the use of machine learning.Assumptions
- Most of what we need to analyze is "time-series forecasting"
- More specifically for sales revenue and inventory level forecasting, we need time-series forecasting with the following attributes: endogenous (vs exogenous), regression (vs classification), don't know between structured vs unstructured, dynamic (vs static), contiguous (vs discontiguous) based on chapter 2 of Deep Learning for Time Series Forecasting.
Sales Revenue Forecasting
- Inputs for sales revenue forecasting:
- date
- value
- event(s) (promotions, etc...)
- Output
- Expected value
Picture of Success
- Be able to create forecast graphs in Metabase similar to this one
- Be able to integrate forecasting into processes like replenishment
- Be able to integrate machine learning into Order Entry => Quick Info tips (decision making) (note: not time-series forecast ML)
Anticipated Project Outputs
- Tables of predicted outputs ($, inventory, etc...)
- Running models that can be queried via API call to get real-time output given a collection of inputs.
Getting Acclimated to Machine Learning
Tooling Options
- Tutorial: Time-Series Forecasting using timescaleDB (open source time-series database that is PostgreSQL compliant)
- AWS General Machine Learning tools
- AWS Time-Series Forecasting (Machine Learning) - only available via preview - waiting on acceptance (2019/01/27)