New ML Methods For Improving Supply Chain Management

We are currently testing the Azure Machine Learning algorithms for demand planning to see what automated ML can bring for the planning community.

One thing that amazed me - how fancy the idea of machine learning can get. It uses new terminology for classic concepts in statistical modeling - Anomaly detection, Training the Model, Testing the Model etc.

Regardless, a good time-series forecast should be part of the planner's toolkit for developing a baseline forecast - a good starting point for the entire demand consensus process with in the S&OP.

Demand Time-series can be divided into three periods:

The model initialization phase is the period to compute a pre-estimation averaging. Most efficient engines may use the entire training phase to compute this average and then re-use this for training the model.

The testing phase is the out-of-sample phase that is a rigorous examination of the quality of the model that is considered the best fit of the patterns in the data. Typically this is the last few periods of the time-series.

Once the algorithm passes this final examination, then it is used to develop the forecast.

Learn more about tuning the algorithms

https://demandplanning.net/workshops.htm