CS2 Cheat Detection using Neural Networks

This project is a proof-of-concept system to detect aimbot-like behavior in CS2 (Counter-Strike 2) matches using machine learning—specifically, an LSTM-based neural network trained on engineered features from in-game demo files.

---

Overview

Goal

To build a neural network that can differentiate between cheater and non-cheater gameplay by analyzing player input data (pitch, yaw, velocity, acceleration, etc.) extracted from .dem files.

Pipeline

1. Parsing

   • .dem files are parsed into CSV using demoparser2.
   • For every kill event, a 300-tick window is extracted for the attacker.

2. Processing

   • Irrelevant columns like player name are dropped.
   • Files are organized by class (cheater vs. legit).

3. Feature Engineering

   • First, second, and third derivatives of aim angles (velocity, acceleration, jerk).
   • Cumulative displacement, statistical summaries, etc.

4. Modeling

   • A two-layer LSTM reads each 300×20 feature segment.
   • Outputs a probability of the segment being a cheater.

---

Results (First Test)

Metric	Value
Accuracy	81%
Legit Recall	91%
Cheat Recall	60%
F1 Score	0.87 (legit), 0.67 (cheat)

These results are based on an initial dataset of 104 labeled segments. Performance is expected to improve significantly as more cheater sessions are collected.

---

Project Structure

The repository implements a real-time cheat-detection pipeline for CS2 using LSTM networks on parsed pitch/yaw data. The main folders and files include:

• data — Data root directory
  • /raw/ — Original .dem files and external datasets.
  • /interim/parsed_csv/ — Parsed tick-by-tick CSVs (before feature engineering).
  • /processed/features/ — Feature-engineered datasets.
• scripts/ — Pipeline scripts to run in sequence:
  1. 00_listSteamid.py
  2. 01_parser.py
  3. 02_process_and_engineer.py
  4. 03_model.py
• notebooks/ — Jupyter notebooks for exploratory data analysis and visualization.
• legacy_ver/ — The first version of the project with a simpler flow and parser, kept for documentation and evolution tracking.

---

Next Steps

• Expand cheater dataset across more sessions and cheat types.
• Tune class weights and decision threshold.
• Explore attention mechanisms and bidirectional LSTMs.
• Improve false-positive control to avoid flagging legit players.

---

Requirements

• Python 3.8+
• Jupyter
• demoparser2
• TensorFlow
• pandas, numpy, tqdm, matplotlib

Installation

Follow the steps below to set up this project locally:

1. Clone the repository

git clone https://github.com/yviler/cs2-cheat-detection.git
cd cs2-cheat-detection

2. Create virtual environment

python3 -m venv venv
source venv/bin/activate

or on Windows

venv\Scripts\activate

3. Install the dependencies

pip install --upgrade pip
pip install -r requirements.txt

Running the Project

1. (Optional) Insert data into data/raw/

2. Execute the data scripts pipeline

python scripts/01_parser.py
python scripts/02_process_and_engineer.py

3. Execute the Model script

python scripts/03_model.py