Note: This article was first published in 2020.
Supervised machine learning is quickly maturing and presents tremendous opportunities to answer questions and gain insights. It is giving decision-makers new confidence in asking more from their data. As always, if you want the right answers for your business, you need to start by asking the right questions. But do your insights really require machine learning?
Machine learning can require more processing power, high hosting costs, and sophisticated skillsets for implementation. Sometimes analytics, heuristics, and/or business intelligence can provide the insights you’re looking for at a better ROI. The five-step approach below can help determine when (and whether) to use supervised machine learning.
Step 1: Identify business needs
As with anything that is technology related, start by pinpointing the business needs. What are the top business questions that the business is currently unable to answer?
Step 2: Identify and acquire data
Insights require data, and machine learning and traditional analytical techniques are more effective when data is abundant. Identify what data is relevant to your problem and assemble it - similar to what you would do for any traditional business intelligence purpose.
Step 3: Make sure all data has the right context
Data is meaningless unless viewed in the proper context. This means cleaning, normalizing, and tickerizing data. This is true for both structured and unstructured data.
Step 4: QA through visualization
Verify that data quality is good. This is best done with simple visualization techniques, such as graphs and charts. This makes it possible to quickly identify outliers and mistakes.
Step 5: Obtain insights via machine learning where applicable
Steps 1-4 are necessary for preparation. This is the point at which it will be clear whether or not you need supervised machine learning.
If you have determined that machine learning is right for you, then be sure to start with the simplest viable algorithms before going after complexity. For example, try a linear regression algorithm before investing in non-linear regression.
Written by Ali Usman, CEO of PixelEdge.
Like what you read?