I’ve recently written about the risks of machine learning (ML), but with this post I wanted to take a step back and talk about ML and general. I want to talk about the ‘why’ of machine learning and whether you and/or your company should be investigating machine learning. Do you need machine learning? Maybe. Maybe not.
The first question you have to ask yourself (and then answer) is this: Why do you want to be involved with machine learning? What problem(s) are you really trying to solve? Are you trying to forecast revenue for next quarter? You can probably do just fine with standard time series modeling techniques. Are you trying to predict house prices in cities/neighborhoods around the world? Machine learning is probably a good idea.
I use this rule of thumb when talking to clients about machine learning:
- If you are trying to forecast something with a small number of values / features – start with standard forecasting / modeling techniques. You can always move on to machine learning after working through the standard approaches.
- If you need to combine multiple data sets to create new knowledge and actionable insights, you probably don’t need machine learning.
- If you have a complex model / algorithm with many features, then machine learning is something to consider.
The key here is ‘complex’.
Sure, machine learning can be applied to simple problems but there’s plenty of other approaches that might be just as good. Take the forecasting revenue example – there are multitudes of time series forecasting techniques you can use to create these forecasts. Even if you have hundreds of product lines, you are most likely using a few ‘features’ to forecast one outcome which can easily be handled by Holt-Winters, ARIMA and other time-series forecasting techniques. You could throw this same problem at a ML algorithm / method and possibly get slightly better (or worse) results but the amount of time and effort to implement an ML approach may be wasted.
Where you get the most value from machine learning is when you have a problem that really vexes you. The problem is so complex that you just don’t know where to start. THAT is when you reach for machine learning.
Do you really need machine learning?
There are a LOT of people that will immediately tell you ‘yes!’ when asked if you should be investigating ML. They are also the people that are trying to sell you ML / AI services and/or platforms. They are the people that have jumped on the band wagon and are chasing the latest buzzwords in the marketplace. In 2 years, those same people will be jumping up and down telling you need to implement whatever is at the top of the buzzword queue at the time. They are the same people that were telling you that you needed to implement a data warehouse and business intelligence platforms in the past. Don’t get me wrong – data warehouses and business intelligence have their places but they weren’t right for every organization and/or every problem.
Do you need machine learning? Maybe.
Do you have complex stream of data that you need to process and turn into knowledge and actionable intelligence? Definitely look into machine learning.
Do you need machine learning? Maybe not.
If you want to ‘do’ machine learning because everyone else is, feel free to investigate it and start building up your skills but don’t throw an enormous budget at it until you know beyond a shadow of a doubt that you need machine learning.
Or you could call me. I can help you figure out if you really need machine learning.