CHEAT DETECTION IN COUNTER-STRIKE 2 USING LONG SHORT- TERM MEMORY NETWORKS

Moehammad Azzriel Ilham, Faizah, S.Kom., M.Kom.

2025 | Skripsi | ILMU KOMPUTER

Penggunaan alat curang yang meluas, seperti aimbot, triggerbot, dan spinbot, merupakan masalah besar bagi pengalaman pemain dalam game first-person shooter (FPS) online yang kompetitif, terutama di Counter-Strike 2 (CS2). Skripsi sarjana ini berfokus pada pendeteksian kecurangan bantuan bidikan dengan menganalisis perilaku membidik pemain yang diekstrak dari rekaman demo gameplay. Parser khusus dikembangkan menggunakan pustaka demoparser2, memungkinkan ekstraksi data gameplay berbasis tick yang mendetail, termasuk pitch, yaw, posisi pemain, jenis senjata, jarak kill, dan kecepatan pemain. Untuk meningkatkan potensi data ini, pembuatan fitur yang komprehensif dilakukan, mengubahnya menjadi 79 fitur berbasis gerakan dan kontekstual yang berbeda. Fitur-fitur tersebut meliputi kecepatan sudut, akselerasi, sentakan, stabilitas bidikan, intensitas jepretan, lompatan posisi, dan detail spesifik senjata. Neural Network Long Short-Term Memory (LSTM) dilatih pada dataset seimbang yang terdiri dari 2212 segmen gameplay berlabel, dibagi menjadi set pelatihan (70%), validasi (15%), dan pengujian (15%) menggunakan pengambilan sampel bertingkat. Model akhir mencapai akurasi 77%, dengan precision dan recall yang seimbang, yang secara efektif meminimalkan false positive dan mengidentifikasi perilaku curang dengan efektif. Early Stopping digunakan untuk mencegah overfitting, yang akan menghentikan pelatihan ketika kehilangan validasi berhenti meningkat. Hasil ini menunjukkan kelayakan dan potensi efektivitas analisis perilaku berbasis pembelajaran mesin untuk deteksi cheat dalam game FPS, meletakkan dasar untuk integrasi di masa depan ke dalam sistem anti-cheat waktu nyata dan alat moderasi dalam lingkungan esports yang kompetitif.

The widespread use of cheating software, such as aimbots, triggerbots, and spinbots, poses a great threat to the player experience in online competitive first- person shooter (FPS) games, particularly in Counter-Strike 2 (CS2). This undergraduate thesis focuses on detecting aim-assistance cheats by analyzing player aiming behavior extracted from gameplay demo recordings. A custom parser was developed using the demoparser2 library, enabling extraction of detailed, tick-based gameplay data, including pitch, yaw, player positions, weapon types, kill distances, and player velocities. To enhance the potential of this data, a comprehensive feature engineering was performed, transforming it into 79 distinct motion-based and contextual features. These include angular velocity, acceleration, jerk, aim stability, snap intensity, positional jumpiness, and weapon-specific details. A Long Short-Term Memory (LSTM) neural network was trained on a balanced dataset comprising of 2212 labeled gameplay segments, split into training (70%), validation(15%), and test(15%) sets using stratified sampling. The final model achieved an accuracy of 77%, with balanced precision and recall, effectively minimizing false positives and reliably identifying cheating behaviors. Early stopping was employed to prevent overfitting, which would stop training when validation loss stopped improving. These results demonstrate the feasibility and potential effectiveness of machine learning-based behavioral analysis for cheat detection in FPS games, laying the foundation for future integration into real-time anti-cheat systems and moderation tools within competitive esports environments.

Kata Kunci : CS2, Cheat Detection, LSTM, Cheating Software, Parsing, Player Behavior