Time-Window Normalized Dataset for Video Game Cheat Detection using Neural Networks

This dataset, contained in the file dataset_Cheat_Detection, consists of 3,584 normalized instances designed for training, validating, and benchmarking machine learning models—specifically deep neural networks—applied to anti-cheat systems in video games. The data represents telemetry metrics and player behaviors extracted and aggregated using a sliding temporal window of 2 seconds. To capture both the behavioral trends and the anomalous peaks of potential cheaters within each window, features were engineered by calculating statistical metrics such as the mean, maximum, minimum, and cumulative sums of the raw game events. The primary objective of this dataset is to allow models to distinguish between legitimate gameplay and multiple concurrent forms of cheating based on short-term behavioral dynamics. Target Categories • es_trampa_max: General binary indicator. 1,0 specifies that a cheat was active during the time window; 0,0 represents legitimate gameplay. • isAimbot_max: Indicates the usage of aiming assists or automatic targeting hacks (1,0 active, 0,0 inactive). • isVidaInfinita_max: Indicates God Mode or health manipulation hacks (1,0 active, 0,0 inactive). • isMunicionInfinita_max: Indicates ammunition freezing or infinite ammo hacks (1,0 active, 0,0 inactive). • isSuperVelocidad_max: Indicates speed hacks or abnormal player movement modifications (1,0 active, 0,0 inactive). Feature Columns The features capture the kinematics, combat stats, and resource states of the player within the 2-second timeframe: • Kinematics & Aiming: angle_error_mean, angle_error_max, speed_mean, speed_max. • Combat Events: is_shot_sum, is_hit_sum, is_received_damage_sum, accuracy_rate. • Resource Status: health_mean, ammo_left_min, ammo_max, infinite_ammo_ratio. • Temporal tracking: timestamp.