Published August 7, 2025 | Version v1
Software Open

Communication-Level Handover Supervision for Fog-Limited WDM-FSO Links: Physics Baselines, Channel-Proxy Audits, and Reproducible QoS Decisions

Authors/Creators

  • 1. AASTMT

Description

Codes and data are updated in the new Zip "Codes and Data Updated.zip" which includes the novel new work of "Communication-Level Handover Supervision for Fog-Limited WDM-FSO Links: Physics Baselines, Channel-Proxy Audits, and Reproducible QoS Decisions"

____________________________________________________________________________________________________

fso_predictor is an open-source, command-line interface (CLI) package for end-to-end data processing, model training, and handover simulation in multi-user, Wavelength Division Multiplexing (WDM) Free-Space Optical (FSO) networks. Written in Python 3.10+, it encapsulates our entire research pipeline—data cleaning, feature engineering (with optional PCHIP interpolation), regression modeling (Random Forest, Gradient Boosting, KNN, Linear Regression, Bayesian Ridge, and optional Neural Nets), hyperparameter tuning via Particle Swarm Optimization (PSO), SHAP-based interpretability, and QoS-driven handover simulation—into a single, reproducible workflow.

Key Features

  • Modular Architecture: Separate subcommands for each stage (preprocess, train, tune, explain, predict, simulate, evaluate) driven entirely by human-readable TOML configuration files.

  • Reproducibility: Locked dependencies, pre-commit hooks, CI/CD badges, Binder demos, and example workflows ensure bit-for-bit replication of our results.

  • High Performance: Random Forest inference is up to 100× faster than comparable neural networks, enabling real-time handover decisions.

  • Interpretability: SHAP summaries reveal that path range dominates link performance predictions—even across SISO and MIMO fog scenarios.

  • Data Transparency: The accompanying 3 000-point FSO dataset is archived on Mendeley Data (doi:10.17632/8jmt7bjn8w.1).

  • Flexible Installation:

    bash
     
    pip install -e .[dev] # Core + dev tools pip install "fso_predictor[interpret,pso,nn]" # Full feature set

  • Extensible CLI: Easily plug in new regressors, interpolation schemes, or deployment scenarios without touching source code.

Usage Example

bash
# 1. Preprocess raw link measurements fso_predictor preprocess --config configs/preprocess.toml # 2. Train Random Forest on Scenario 1 fso_predictor train --config configs/train_rf_sce1.toml # 3. Tune RF hyperparameters via PSO fso_predictor tune --config configs/tune_pso.toml # 4. Generate SHAP summary plots fso_predictor explain --config configs/shap.toml # 5. Simulate QoS-driven handover along a user mobility trace fso_predictor simulate --config configs/simulate.toml

Citations & Links

Files

Codes and Data Updated.zip

Files (13.0 MB)

Name Size Download all
md5:3be7f5acb94ee9ba710e2965c55589bc
245.4 kB Preview Download
md5:72ab3276887c54ed79f5f28998204955
91.0 kB Preview Download
md5:1e0bb2543bcc9ffb3c5e8d865b7ee0c0
12.7 MB Preview Download

Additional details

Software

Repository URL
https://github.com/AIBabyTeaching/PredictiveModellingRebirth
Programming language
Python
Development Status
Unsupported