I have recently introduced an approach on how to frame Machine Learning (ML) solutions. In short, you should iterate a three-step process that includes refining the problem, understanding the data requirements, and exploring the baseline solution. In this article, I am sharing a practical approach to do so. The process, I hope, would be less fuzzy than having a light-bulb idea and figuring out where it fits, as illustrated in Figure 1.
A machine learning solution in its core is a code that reads data to generate a model and then uses another set of data from the same source to generate predictions. If you follow a pragmatic approach, like the three-step process, it is unlikely that the first draft solution would stick in the end. Rather, as depicted in Figure 2, you are likely to evolve the solution several times wherein each iteration you are searching for better data and/or developing an improved code. The question is how many drafts you should work towards.
Drawing inspiration from my engineering and research background, I advocate following six steps to make the process a bit more efficient and practical (see Figure 3).
- Define an impactful business goal together with your stakeholders
- Set a reasonable deadline together with your stakeholders
- Calculate a baseline performance
- Build a simple, fast, improved solution
- Improve your solution until you achieve the baseline performance within your deadline by iterating step 4
- Decide next steps with your stakeholder upon reaching the deadline
We will provide more details on each of these steps.
Defining business goal
The first thing you should do is make a business case and a reasonable goal. If your stakeholders are not convinced about your goal, you need to revisit the goal and/or the case. Avoid being too ambitious, since it is better to learn to walk before running.
Tie the goal to a business metric, e.g., sell-through, churn rate, etc. The former metric indicates the percentage of products sold by a business from its total stock over a period and the latter indicates the fraction of the customer that has stopped being an active customer after a period. Obviously, the higher the sell-through or the lower the churn rate, the…
Continue reading: https://towardsdatascience.com/practical-method-to-framing-machine-learning-solutions-8b36a9d40efb?source=rss—-7f60cf5620c9—4