👨‍💻 Data Scientist | ML Engineer | Digital Twins | Data Analytics

I specialize in applied machine learning, deep learning, explainable AI, and digital twins for industrial applications. My focus is on bridging the gap between complex ML models and real-world business solutions.

I am actively involved in research on explainability in reinforcement learning and digital twins. You can find my publications here:

🔹 Demystifying Reinforcement Learning – SHAP and Captum
Exploring explainability in reinforcement learning using SHAP and Captum to make black-box models more interpretable.

🔹 Learning Simulation-Based Digital Twins for Discrete Material Flow Systems
Systematic Literature Review shedding light on Digital Twins for discrete material flow systems.

📚 Full Repository: Scientific Papers Repository

Description: A PyTorch-based implementation for interpreting neural network predictions using Captum.

• Utilizes Integrated Gradients, SHAP, and Feature Attribution to explain model decisions.
• Custom implementation for multiclass classification explainability, overcoming Captum's default limitations.
• Example notebooks demonstrating explainability for different architectures.

Tech: Python, PyTorch, Captum

📌 How to Use: Clone the repo and run the notebooks to visualize model explanations.

Description: An automated validation framework for digital twins in discrete material flow systems, supporting Industry 4.0 applications.

• Innovative validation pipeline ensuring digital twin accuracy in production environments.
• Adaptive xLSTM algorithms for real-time anomaly detection (75% accuracy).
• Configurable metrics dashboard for performance monitoring.
• Validation testing suite for simulation fidelity across multiple parameters.

Tech: Python, PyTorch, Docker, PostgreSQL, Pandas, Matplotlib

📌 How to Use: Clone the repo, navigate to the val folder.

Description: Award-winning soil moisture prediction model for smart irrigation.

🏆 1st Place Winner – Open Innovation City @ Founders Foundation Bielefeld

• Predictive modeling for soil moisture with 78% accuracy.
• Gradient boosting implementation optimized for changing weather conditions.

Tech: Python, Scikit-Learn

Description: YOLOv8-powered model for detecting enemy soldiers in Call of Duty: Warzone. Exploring computer vision for gameplay analysis & cheat detection.

• RQ1: How well does YOLOv8 detect enemy silhouettes in FPS games?
• RQ2: Can CV models help identify potential cheats?

• Detection Mode: Warzone dataset (labels: "enemy", "head")
• Pose Mode: Zero-shot silhouette detection

• Warzone images (Roboflow, CC BY 4.0)
• Custom test set (newer game versions)

• Hardware: RTX 4070 Ti, CUDA 11.8
• Software: PyTorch 2.3.1, AdamW (LR: 0.01)
• Training: Early stopping (3 epochs patience)

• mAP50-95: 0.530 (object detection)
• Pose model excels in silhouette detection
• Generalizes to new game versions
• Struggles with occlusions (e.g., head-glitching)

• Esports analysis
• CV model training data
• Virtual environments & metaverse

• Multi-modal fusion (visual + audio)
• Enhanced occlusion handling
• Cross-game transfer learning

Description: A collection of my Kaggle competition projects, solutions, and data science approaches.

• Competition solutions with detailed explanations and methodologies.
• Exploratory data analysis notebooks showcasing my analytical approach.
• Model implementations with performance metrics and comparisons.
• Feature engineering techniques that improved model performance.

Tech: Python, PyTorch, Scikit-Learn, XGBoost, Pandas, NumPy

💼 LinkedIn: linkedin.com/in/danielfischerbielefeld

📂 Xing: xing.com/profile/Daniel_Fischer635