17 March 2017 | admin

Machine Learning Masterclass #7: What can Machine Learning teach us?

In our Machine Learning Masterclass series so far, we have covered what Machine Learning actually is, the impact it might have on SEO and the sorts of tools and applications that are currently powered by machine learning. In this blog post, we will explore the principles of Machine Learning that we can be applying to our current data analysis practices.

Before Machine Learning is applied to run awesome prediction algorithms and discovering insights, there is a lot of preparation that goes into it. It is these preparations that make Machine Learning such a reliable data analysis methodology.

Define the problem better

The first thing to learn from Machine Learning is how to formally define a problem.

Problem definition is one of the very first and perhaps the most crucial step in rolling out Machine Learning. After all, there is no point in using the most powerful and clever algorithms if you are solving the wrong problem. In Machine Learning, the problem definition goes through a stage of informal definition first and then it is formally documented.

Learning from this, digital marketers can get into the habit of transforming the clients’ business objectives and statements into a structured and a documented measurement plan. The Google Analytics Academy covers this topic in great detail here.

 

Clean your data

The second crucial point Machine Learning teaches us is that it is paramount that the data you are analysing has been ‘cleaned’. Tidy data in Machine Learning allows easy manipulation, modelling and visualisation.

Ensure that you have organised your analytics account in a structured manner with filters and custom dimensions etc.

Using account structuring in Google analytics you can:

  • Avoid including your internal staff traffic from skewing your analysis
  • Exclude any spam traffic from being reported, and
  • Exclude traffic figures from pages that are the result of an incorrect setting in your CMS.

 

Example of filters in Google Analytics

Filters in Google Analytics

Label your data as much as you can

Labelling data is perhaps the most useful cue to take from Machine Learning. Labelling is so crucial to Machine Learning that there are algorithms to label the data itself. This makes sense for Machine Learning as sometimes labelling has to be applied to data with millions of rows. This article explains how data labelling could be the next blue collar job once AI and chat bots take over.

Labelling data can be easily practised and applied within your current datasets. This will allow you to quickly organise your data into groups and clusters that will uncover some interesting insights.

For example, in your keyword ranking report you can label all your local keywords as local+region and monitor ranking movements of similarly labelled keywords by regions.

Example of data labelling on keyword report

 

Leave room for error prediction

Quite often in Digital Marketing you are required to put together predictions and  forecasts. Machine Learning, can help with this although be aware that even the best predictor will sometimes be wrong. Presenting this prediction error is a key facet of Machine Learning, and is worth utilising in your current data analysis practices.

Make it visual

Lastly, no Machine Learning presentation or talk is complete without a mention of visualisation tools and practises. Whilst Machine Learning algorithms allow data scientists to run complex algorithms on vast data sets, sometimes their outcomes and findings are fairly cryptic and comprehensible only to the subject matter experts. To resolve this data scientists rely on visualisation tools.

This is a key take away for any digital marketing analyst. It’s great if you can collect large volumes of data, run analysis and derive insights, representing it in an easy to digest format.

Graphical representation of a classification algorithm in form of a decision tree

Graphical representation of a classification algorithm in form of a decision tree

 

It is understood that a lot of detailed planning and investment is required in rolling out Machine Learning practises in your organisation. However, it does not mean that you should not apply some of the principles of the field in your current data analysis techniques. This blog attempts to highlight some of the Machine Learning best practises that can significantly improve your data analysis capabilities.

Feel free to contact one of our experts to discuss this in more detail.

admin