How danielzlln/bachelorthesis_bs works
Project Type and Purpose
• This is a data‐science‐driven dashboard application (“Bachelorthesis”) whose goal is to forecast bicycle traffic in Münster, evaluate model performance and explainability, and present results via a local web dashboard.
High‐Level Components
a. Data Sources
• Raw traffic counts (XLSX files under src/)
• External CSV ingest (raw_github_data)
b. Data Processing Layer (data/)
• Cleaning scripts (clean_data.py, imputation.ipynb)
• Feature engineering (feature_engineering.py, Ferien_Feiertage calendar)
c. Model Training & Artifacts (notebooks/)
• Modeling notebooks for Prophet, LSTM, GRU, Random Forest, XGBoost
• Stored model artifacts and metadata (pickled scalers, trained models, best‐params CSVs, experiment logs)
d. Dashboard Backend (application.py)
• Flask or Dash application entry point
• Loads cleaned data, features, and serialized model artifacts
• Exposes REST endpoints (if any) and serves HTML/JS/CSS
e. Frontend (assets/)
• style.css, custom layout or Dash components
f. Presentation/Reports (reports/)
• Static report modules (optional)
Relationships & Data Flow
1. Raw Data → Data Processing
2. Cleaned Data → Model Training
3. Model Artifacts + Cleaned Data → Dashboard
4. Dashboard → User
Architectural Patterns & Design Principles
• Layered architecture separating:
• Data ingestion/processing
• Model training & artifact management
• Application server (dashboard backend)
• Presentation layer (UI assets)
• Reusable modules (data/ and feature_engineering.py can be invoked by both notebooks and application.py)
• Artifacts versioning: notebooks/model_metadata stores parameters and pickles for reproducibility