# Can I use a clustering algorithm or some mechanism to classify parts?

I am new to the site if I am posting on the wrong site sorry about that and please if so lead me to where I can post this...

I work in the automotive industry but specifically an area that deals with packaging, logistics, and the most efficient ways to ship parts from a supplier to an automotive plant (so that cars can be built).

We have a term we use called "Commodities" that is used to define a grouping of parts. For instance, there are vehicles that use the same type of rear view mirror and we will usually group this into the same commodity...BUT...

There are certain commodities that when you read them on paper seem the same but as you dig deeper and look at the characteristics of the part you notice it cannot be grouped into the same commodity. One of those factors are weight...because the weight is so much different we cannot lump commodity X with commodity Y because commodity y's weight is 5 times that of X. Can you imagine the car wire harness for a small car vs that of a wire harness for a truck?

They are both "wire harnesses" but the wire harness for the truck weighs much more. On paper both parts may look like this:

Part Number  Part Description   Weight
123456       Wire-Harness        3
123777       Wire-Harness        12
124444       Wire-Harness        12.18
123222       Wire-Harness        12.34


If we looked just at the part description we'd have one commodity called "Wire Harness". However, since the weight plays a big factor (you cannot put in the same number of wire harnesses in a box to ship if their weight is that much off) we end up having to split this up into another commodity. In our data above the last three parts could possibly be in one commodity while the first part could be in its own commodity.

So on to my question...can I mathematically solve this using some sort of algorithm in which I can plot some data and see a trend on what should be seperated based on weight? Can I use some sort of clustering algorithm:

https://towardsdatascience.com/the-5-clustering-algorithms-data-scientists-need-to-know-a36d136ef68

And plot out, for instance using my example of a wire harness, all the weights we have for a wire harness and see what this leads me to in terms of how many commodities? If there is a way to do this mathematically can someone assist with how this may look or how I can work this out?

• Maybe Cross Validated is a better home for this question? – kimchi lover Nov 8 at 16:10
• How you do this depends on why you want to separate harnesses by weight. For instance, if the problem is carrying requirements, you can divide them into how many workers would be required (a problem we have with aircraft harnesses, though maybe cars can be done without lugging them in by hand?). Where you make the divisions should be based on why you want to make the divisions. – Paul Sinclair Nov 9 at 4:22