ML Use Case — Part 1

In this age of hyper competition, machine learning use cases hold the key to success in the market. Oh well, you must have heard that statement multiple times, hyper competition, super regional, micro focused etc. I don’t even know and remember the various combinations ‘these’ words any mover. However there is a good amount of truth in it.

Let us take few example that one encounters on regular basis and have some discussion around it.

One can’t make a larger shirt with less material. Additional material should increase the price. Apparently, it is not so with Dockers shirt as shown in adjacent image. Medium size of ‘Ultimate Button Up Smart 360 Flex Shirt’ is $36.99, while the Large size of the same variety is $12.50. Almost one third of the smaller size. One might have noticed, I did not call this ‘discrepancy’, as it could be by design. Well, there could be several explanations to this. Few explanation that I can immediately think of — Docker makes one large shirt of the remaining material from 3 medium shirts. There is demand for 30 medium shirts, however the demand for large shirts is 3. Hence the reason for that one third shirt price. Another explanations — those are the last three pieces of large shirts left, and Dockers has recovered cost and margin on that lot, new lot is coming out soon, etc. However one gets the idea.

Take a look at another example from IZOD. One can observe the price variation is phenomenal, ranging from $10.84 to $77.97. I guess, I should be able to find similar variations to docker, if I inspect the prices of each size and color. That is not the point I want to make here. The point is — look at variations. There are 28 varieties in 5 sizes. This is relatively simplistic variation I have chosen, there are combinations of 40+ color/patters with sizes such Large — Regular, Large — Slim, Large — Tall, Large — Tall — Slim, X, XX, 3X, Big — 2XL etc. The problem will be compounded when supply chain across the globe is considered, for example prices in US will be different than prices in Europe. Don’t forget the sizes as well. US’s Medium is Europe’s Large — is that right? Currency conversion, Local seasonality will add to additional layer of complexity.

BTW — I am big fan of both the brands Dockers and IZOD.

If someone is still toying with idea of maintaining a spreadsheet with all the combination of merchandise along with prices, discounts, commissions etc. best luck with that.

There are many business who still do this, and here is a sample where different variation sizes, product lines etc is maintained in great details and sent across to agents on regular basis, the headers considered — # of pieces, MRP, Retailer Margin, Retailers Base Price, Scheme (if available) discount, Net Rate, Taxes, Stockist/Wholesalers Margin. Now combine that with daily changes, tracking, seasonal/regional trend discoveries. The constant challenge of maintaining this kind of system explains why I wished luck to owner of this system.

As this can’t be done manually, to predict and set the prices one will need help of some form or variation of machine learning. This brings to the original statement, In this hyper competitive …

There are plenty of use cases, Few of them are listed below to give an idea:

1. Predict the most profitable price for this combination of CPU, Memory, Storage and Network; aka Virtual Machine on any of the Cloud Platform
2. Discover at what temperature range, humidity range, dust particle levels and light range the yield for integrated circuit will be maximum
3. What combination of purchase item, purchase location, purchase quantity, range of variations, past pattern indicate this purchase could be a fraud
4. Should I buy this stock given the price variation in the past, along with book value, market volume, debt ratio, competitive positioning etc.
5. What could be the next best action such as TV show recommendation OR purchase recommendation can be given to user on my web site.

All these and hundreds of use cases out there can be addressed easily with relatively simple implementation of machine learning.

Identifying a use case is up to business and subject matter expert. Searching on web and business background along with prior experience will certainly help to form a good business use case. This is where journey to develop an algorithm starts.

Machine Learning — Definition

To borrow & slightly modify the definition from the wikipedia, Machine Learning is the scientific study of algorithms and statistical models that computer systems use to perform a given business use case without using explicit instructions, relying on patterns and inferences found within the the training data.

There are various ways to implement these machine learning algorithms. We’ll visit one of the easiest in my next article ML Use Case — Part II.