Classification Modelling with case Sortter
0 likes366 views
The document discusses classification modeling in machine learning, focusing on predicting outcomes based on input variables, such as website user conversions. It emphasizes the importance of data collection and cleaning, utilizing tools like Google Tag Manager and Supermetrics for better analytics insights. The presentation highlights that understanding analytics and statistics is essential for applying machine learning without the need for coding.
Classification Modelling with case Sortter
- 1. Classification Modelling with case Sortter #mcsthlm19 – October 26, 2019 @mertanen
- 2. Short history Petri Mertanen • https://www.linkedin.com/in/petrimertanen/ • BBA (Marketing), Specialist Qualification in Management • (Digital) Analytics exprerience since 2005, lecturer at Aalto University 2017-2018 • Certifications: • Elements of Artificial Intelligence: https://www.elementsofai.com/ • Statistical thinking for Data Science and Analytics: https://www.edx.org/course/statistical-thinking-for-data-science-and-analytics • Google Analytics Individual Qualification, Google Ads Fundamentals, Google Tag Manager Fundamentals, Introduction to Data Studio • MeasureCamp Amsterdam 2017: Predictive Conversion Modelling • MeasureCamp Amsterdam 2018: From Digital Attribution to Marketing Mix Modelling • MeasureCamp Amsterdam 2019: Linear Regression Analysis and Modelling
- 4. About classification modelling • In Machine Learning and statistics, classification is the process of predicting the certain output from input variables. • For example, we can predict if the website user / visitor will convert or not? • Classification is an example of pattern recognition. • In Machine Learning, classification is considered an instance of supervised learning. • Learning where correctly identified observations (outputs) are available. • Analyzing which variables or features explain the certain output? • For example, we can analyze which content, elements or features explain the conversions • Variables can be categorical or numerical • With case Sortter, we can predict and analyze things related to loan applications • Decision tree is one of the used classification algorithms • https://en.wikipedia.org/wiki/Statistical_classification
- 5. Collecting and cleaning data • By default, current Google Analytics tracking is crap… • It doesn’t track outbound links, videos, email or phone clicks, forms, file downloads, scrolling, transactions, unique sessions…or other elements / features in website or app. • Luckily, we have Google Tag Manager…and yes, GA v2 is available • If you want to know which content, elements and features are affecting on conversions, and how much, you need to track these in the first place. • Instead of aggregated data, save more detailed data with session, client or user IDs • Useful custom dimensions for Google Analytics: https://www.simoahava.com/analytics/13-useful-custom-dimensions-for-google-analytics/ • For example, you can use Google Analytics APIs to pull the data out • Make sure that your data is clean and in Machine Learning ready format
- 7. Data preparation • Major “healthcare” player in Finland. • 46 % of bookings come from online. • Mostly used paid search (Google Ads), Facebook and display as digital advertising channels. • Paid search being clearly the biggest channel according to money spent. • Data collected during Q1/2019 on daily level. • You can do the data cleaning manually in Excel. • Or you can automate the data collection, for example with Supermetrics (from Finland!).
- 8. Platforms for AI and ML
- 9. Different options to connect data source(s)
- 10. 1-click dataset
- 13. 1-click or configured Linear Regression
- 19. Summary and key takeaways • You don’t need to code in order to practice Machine Learning and Data Science • But you have to understand Analytics and Statistics • Classification modelling can be a solution if the output is kind of “yes / no” • There are other modelling options as well • You can analyze different kind of things which explain the output • If you analyze website content, elements and features, you will get nice insight on where to do A/B and Multivariate testing
- 20. Questions? Petri Mertanen Analytics Consultant petri@mertanen.info Mobile: +358 400 792 616 Social media Linkedin.com/in/petrimertanen/ Twitter.com/mertanen Slideshare.net/mertanen