Predicting the Probability of a Click on the Event Using Machine Learning Methods

Summarizing this work, it is important to note once again that the growing number of Internet audiences influences the business of ticket operators, in particular, the number of tickets sold with them, and the availability of an effective IT solution makes it possible to extract from this influence a key advantage - increasing incomes by selling more tickets. This can be achieved by predicting the likelihood that the user will click on events that are offered in the SERP. In the process of carrying out the research, a real-time analysis of the company Ticketland carried out a domain analysis and a review of the application of the technology of machine learning. Also, the existing algorithms that allow the prediction of the probability of a click on an event were considered in detail. Taking into account the subject area and the features of the data, the criterion for assessing the quality of the classifiers' work was chosen, namely the AUC-ROC metric. Within the framework of practical implementation, the company collected raw data, analyzed and pretreated it. On the basis of the analysis and processing, a sample was formed for the training of classifiers, the construction of which constitutes the main part of the work. As a result of the work, the main goal of the research was achieved: the construction of an effective model capable of predicting the probability of a click on an event that is offered to a user on the site, and the computation of informative signs that affect the user's decision. Such a model based on the analysis is a gradient boost model that shows a classification quality of ~ 87%.