Titelaufnahme
Titelaufnahme
- TitelPlayer profiling in CS:GO: leveraging machine learning for mouse movement-based identification / Lars Gansel ; supervisors: Prof. Dr. Patricia Arias Cabarcos, Dr.-Ing. Philipp Terhörst
- Autor
- Gutachter
- Erschienen
- Umfang1 Online-Ressource (v, 60, 14 ungezählte Seiten) : Illustrationen, Diagramme
- HochschulschriftUniversität Paderborn, Masterarbeit, 2024
- AnmerkungTag der Abgabe: 17.07.2024
- Datum der Abgabe17.7.2024
- SpracheEnglisch
- DokumenttypMasterarbeit
- Schlagwörter (GND)
- URN
- DOI
Links
- Social MediaShare
- Nachweis
- IIIF
Dateien
Klassifikation
Abstract
The prevalence of smurfing and boosting in online games, wherein skilled players either play on lower-skilled accounts or artificially inflate account ratings, undermines fair matchmaking and diminishes player enjoyment. Since players do not spent money on games which they do not enjoy, developers are incentivized to stop smurfs and boosters. This thesis addresses the need for effective detection of such practices, focusing specifically on the game Counter-Strike: Global Offensive (CS:GO). By utilizing a unique data set derived from publicly accessible CS:GO demos, this research aims to assess the effectiveness of using mouse movement data to verify and identify player identities. Two research questions guide this study: “How effective are mouse movements for verifying player identities in the game CS:GO?” and “How effective are mouse movements for identifying player identities in the game CS:GO?”. To answer these, Deep Learning (DL) models were developed and trained using mouse movement data extracted from professional CS:GO matches which contain 69 years of continuous mouse movement. The verification model achieved an average F1 score of up to 0.957 (±0.026) and an average Equal Error Rate (EER) of 0.036 (±0.020). The identification model achieved an average F1 score of up to 0.941 (±0.051) and an average EER of 0.009 (±0.009). When sequences from a single player and match were grouped, the models achieved an average F1 score of 0.990 (±0.045) and an average EER of 0.001 (±0.004). Although the proposed models demonstrate significant improvements over the current state of research, further refinement is essential for deploying them in real-world applications to autonomously detect smurfs and boosters. Nevertheless, these models could serve as an initial filter to flag suspicious activity for manual review. The approach and data set of this study can be utilized for future research to advance player recognition techniques using behavioral biometrics, potentially extending beyond CS:GO to other games and applications.
Inhalt
