How to find excellent Machine Learning Solutions

You’re thinking about trying a machine learning solution for your company, but don’t really know how to tell good machine learning from bad. In this golden age of data, intelligent data usage has become the difference between success and failure for many companies. Machine learning is this decade’s hottest approach to problem-solving. Many businesses are looking for machine learning solutions to remain competitive, but have little knowledge of what to look for in potential hires. They often struggle to decide between statistical solutions and machine learning solutions.
Understanding the key conceptual distinction between statistical methods and machine learning should help you decide on which approach would most effectively address your business problem.
The normal stats offering
Statistics usually asks ‘what’ and sometimes ‘why’. It is descriptive and explanatory. It explores what is happening, the reasons why things are happening (correlation), and sometimes the causes as well. A statistician will find out how many people are regularly downloading apps in Nigeria (what the reality is) and will use statistical methods like regression to investigate probable causes for rates of app download (why the reality is the case). One could find that falling data costs and availability of cheaper smartphones than before is influencing an increase.
But a statistician’s focus is initially historical. If you want to predict whether the rate of app downloads will change in future, statistical methods are less powerful, because they follow from an inferential focus and not a predictive focus. Statistical methods are primed to find an inference.

Machine learning is focused differently
Machine learning methods are primed to predict. The core focus is ‘what will happen next’. Machine learning develops algorithms to gauge how situations will likely develop, and refines the algorithms over and over until the predictions are sufficiently accurate. It is usually more effective at predictions than statistics because it is better at integrating thousands if not millions of potential influences. But machine learning is often less useful than statistics at explaining why a reality is the way it is.
Importantly, machine learning developers can provide demonstrations of success. Developers can show that their algorithms yield sufficiently accurate results. You would never be able to get a statistics-based consumer research house to prove the accuracy of their results in this way, because the only way to truly validate the outcome would be to replicate the study. But with machine learning, you can. So if you’re hiring, ask developers to show you that their algorithms predict with an impressive degree of accuracy.
Depending on the business problem, the format of this demonstration might require tweaking. There is a shortage of easily generalizable machine learning algorithms, and success is often difficult to determine in advance. But, be it a minimum viable product, a test case, contingency fee structure, or even just examples of previous work, machine learning solutions should make predictions of some form. And at some stage, those predictions can and should be tested, and you can ask to see results when considering a machine learning proposal.
Like our ideas? Get the best fresh to your inbox and keep up them on Twitter and LinkedIn.

