How can we apply analytical techniques from other sectors in marketing? (Part 2)
Over the month of November Brightblue R&D team have been looking at analytical techniques from other industries and how they can be applied within the Marketing industry.
We will share with you 3 of those techniques that can be used to provide the most amount of value from businesses data.
Part Two: Classification Methods
We now move on to analytical techniques that fall under the category of classification methods. Classification is slightly different to regression. Although the set up of the model is very similar, the goals of the two models is where the difference lies. Whilst the dependent variable of a regression spits out a prediction of a quantity, the goal of a classification is to predict the correct category of an observation. Let’s take a look at a famous example used by those studying Machine Learning – The Iris Dataset.
Suppose, for simplicity, that we have three types of flowers to ever exist that go by the names of, Setosa, Versicolor and Virginica. If we take a sample of 10 flowers of each, record the lengths and widths and in turn assign the flower to the correct species as shown in Fig 1, the model is then able to recognise certain features and characteristics of each flower
Figures 2 and 3
As a result, classification methods can be used to predict the correct category of new observations based on their characteristics as shown in Fig 2 and Fig 3. If you were to stumble across a random flower on the street and which left you unsure as to which species the flower belonged to, you would be able to simply input the flowers characteristics into your model which would in turn predict which category the observation belongs to.
Relevance in Marketing
So what relevance does flowers and petal lengths have in Marketing? Living in the digital age, we are now able to access so much data that allow us to learn about each individual customer. Using certain metrics such as time spent on website, amount spent per period or other relevant variables, we would be able to segment different type of customers by activity. We would be able to filter out active, moderately active and inactive users. In turn, if a relatively new user was to use the website, by compiling information about them and imputing it into the model, we would be able to allocate them to a category. Consequently, that would allow one to use effective campaigns that are targeted to a more precise demographic as opposed to using a generic advert for a group of customers with a wide range of difference in preference.