Predviđanje kraja interesa korisnika za igru putem strojnog učenja

undergraduate thesis

• Title (english): Predicting user churn in a game by using machine learning
• Author: Krešimir Bačić
• Mentor: Mirko Sužnjević (mentor)
• Committee member: Mirko Sužnjević (predsjednik povjerenstva)
• Committee member: Irena Oršolić (član povjerenstva)
• Committee member: Demijan Grgić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-03, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Podaci o zabilježenim događajima korisnika jedne free to play mobilne igre dohvaćaju se iz lokalne MongoDB baze i prolaze postupak pretprocesiranja tijekom kojeg se pokušavaju zaobići nedostaci skupa podataka. Podaci obrađeni u pretprocesiranju koriste se za izgradnju korisničkog profila - prethodno definiranog seta karakteristika značajnih za strojno učenje kojima se opisuje svaki korisnik. Izabrani korisnički profil sastoji se od broja sjednica, prosječne duljine sjednice, prosječnog broja događaja po sjednici, brojeva otključanih, započetih, prekinutih i uspješno riješenih poglavlja, broja otključanih i broja riješenih priča, pogledanih videa, neuspjelih verifikacija, broja neuspješnih kupovina zbog manjka resursa te količinu potrošenog novca za svaki od 3 perioda: od početka igranja do 72. sata, od 21.6 do 72 i od 50.4 do 72 sata od prvog zabilježenog događaja. Korisničkim se profilima dodjeljuje i klasa, a 80% ih se koristi za treniranje modela dok preostalih 20% služi za evaluaciju. Prije početka treniranja podaci se skaliraju min-maks algoritmom s rasponom [0, 1], a koristi se algoritam strojnog učenja slučajne šume. Rezultati se provjeravaju 30-strukom unakrsnom validacijom.

Abstract (english)

User data collected from the userbase of a free to play mobile game is retrieved from a local MongoDB database and is preprocessed in an effort to circumvent the dataset's imperfections. Preprocessed data is then used in building user profiles – predefined sets of properties relevant to machine learning, each describing a single user. The user profile template used in this thesis consists of session count, average session length, average event count per session, chapter unlock, start, drop and completion counts, story unlock and completion counts, number of promotional videos watched, failed verification count, number of times the player didn't have enough resources to complete a purchase and money spent for each of the 3 observed periods: between start of the first play session and the 72nd hour, between 21.6 and 72 hours and between 50.4 and 72 hours of playtime. User profiles are assigned a class and 80% are used in training the machine learning model, while the remaining 20% is left for model evaluation. Before training the model on calculated profiles, the data is scaled using the min-max algorithm with a range of [0, 1], while random forest is the chosen machine learning algorithm. Results are evaluated using 30-fold cross-validation.

• Keywords: strojno učenje; predviđanje; klasifikacija; algoritam slučajne šume; min-maks skaliranje; MongoDB; pretprocesiranje; umjetna inteligencija; Python; scikit; sklearn
• Keywords (english): machine learning; prediction; classification; random forest algorithm; min-max scaling; MongoDB; preprocessing; artificial intelligence; Python; scikit; sklearn
• Language: croatian
• URN:NBN: urn:nbn:hr:168:468943
• Study programme: Title: Computing; Study programme type: university; Study level: undergraduate; Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 14:24:19