I am currently working on a project to automate parts of warehouse operations, specifically batched order release.
I was wondering someone of you has an idea for an approach in Knime and which nodes to use. I thought of Logistic Regression (as in order release yes or no). I want to take several variables into account such as cutoff time, country, similar orders. As this is a new approach, there is no training dataset but I do have data on orders and warehouse operations and all kinds of analytics and forecasts already.
Does someone know how I could approach it?
Thanks in advance!
Hi @_Karo97 -
You mention there is no training dataset available, but it would be useful to see the kind of data you DO plan to use. Are there time series aspects to this, or is this just classification based on various batch characteristics? I think we need more information to help formulate a solution here.
There is a dataset available. It shows past sales data and picking times. However, we do not have labels of when data has been released before or which batches are good.
I would want to classify the released orders on the following characteristics: time of the day, cutoff of time (pick up time), forecasted demand and resource availability, similarity of orders (such as all shampoo orders together or all clothing orders together) to increase the batch effect. I already analysed the data, made a forecast model and applied association rules. So this information is also available.
Could you perhaps post a sample of your data, along with what your desired output would look like? I apologize, I’m just not very familiar with this type of warehousing / batched order release problem, and I’m having a hard time understanding what you want do to, especially given that you already have forecasting and association rules built out.
You mention logistic regression for classification, but what are you trying to predict? As you have no training dataset yet, do you first want to do some sort of clustering analysis to try and identify possible classes, and THEN classify, or…?
We are considering an order release model for warehouse operations. That means online orders arrive in our system every other minutes. These orders needed to be picked in a warehouse and be fulfilled at a given time each day. When an order arrives, I want a system to say we release it for picking or we wait. This decision should happen based on some attributes such as time of the day, number of orders, capacity, time until pick up but also similarity of orders. Instead of releasing orders as they arrive, we can bundle them. E.g. a shampoo order arrives but until the end of the hour, you expect 5 more shampoo orders. Thus, you would rather not release but wait for one hour and pick these orders together. This saves times (and cost) because the picker has to go to the aisle of shampoo just once in an hour instead of 6 times.
Important here is that we have a diversity of orders and not always the same kind of order.
Logistic regression has a binary outcome which is needed here “yes, release order now” or “no, release later”
This data includes order data, when the order was dropped, which product, when it has to be shipped etc.
Joined tables test.xlsx (11.2 KB)
Unless I’m misunderstanding you - entirely possible - you will need to label your data in some way first before using a classification method like logistic regression.
What are the rules that your pickers would apply if they had to make the decisions manually? Time since order arrival, groupings of similar items, number of orders, customer priority… these are things that your picker would take into account without the aid of an algorithm. So if you can codify these rules, you can then use the rules to apply ship now / don’t ship yet “labels” - which you could then in turn use to build a model.
This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.