User Behavior Analysis for User Profile Prediction

The importance of user behavior analysis is becoming increasingly valuable to Software as a Service (SaaS) businesses such as Kpler. Users of big data applications require attention in order not to lose customer loyalty in a competitive business environment. With data analytics techniques, knowledge can be extracted about the user behaviors, and that would be a beneficial exercise to the business when it comes to improving services to the customers whether it’s a tailored customer support, targeted marketing campaigns or enhanced product features. This capstone project is focused on predicting user profiles through their web usage behavior. A data mining approach was adopted to reach the objective of this project, and interesting patterns were derived from the analysis of the usage data. The main profiles of Kpler users are as follows: trader, operator and analyst. There are two different angles to the usage on the platform, Map and Analytics. It was found that the map feature on the platform was the most frequently used by all user profiles due to the richness of the information presented on the map. The analytics feature of the platform is frequently being used by trader and analyst profiles since they are user-friendly and can analyze trends fairly quickly. Because these patterns can change over time, predicting user profiles can become interesting in evaluating their usage. We’ve selected Random Forest as the classification model, which resulted with the following metrics: train accuracy 96.9%, test accuracy 49.5%, precision: 97%, recall 96% and f-1 score 97%. These scores are considered to be a good starting point for a promising future work on this topic for the company