An Explainable Stacked Ensemble Framework for Multi-Horizon Temperature Prediction in IoT-Enabled Medical Cold Chain Systems
Overview This repository provides the source code, experimental configurations, and supporting reproducibility materials for the research study titled "An Explainable Stacked Ensemble Framework for Multi-Horizon Temperature Prediction in IoT-Enabled Medical Cold Chain Systems." The study investigates OPTIML, a multi-horizon stacked ensemble framework for indoor temperature forecasting using environmental measurements collected from an IoT-enabled medical cold-chain monitoring system. The proposed OPTIML framework integrates ten machine learning and deep learning base learners with a Lasso-based meta-learner to generate temperature forecasts a
Resource information
- Format
- zip
- Original repository
- 23970862
- DOI
- 10.5281/zenodo.23970862
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- cc-by-4.0
- Last record update
- Nov 5, 2026
About this resource
Overview This repository provides the source code, experimental configurations, and supporting reproducibility materials for the research study titled "An Explainable Stacked Ensemble Framework for Multi-Horizon Temperature Prediction in IoT-Enabled Medical Cold Chain Systems." The study investigates OPTIML, a multi-horizon stacked ensemble framework for indoor temperature forecasting using environmental measurements collected from an IoT-enabled medical cold-chain monitoring system. The proposed OPTIML framework integrates ten machine learning and deep learning base learners with a Lasso-based meta-learner to generate temperature forecasts across four future observation horizons (t+1 to t+4). The study also investigates an exploratory persistence-anchored robust residual extension, referred to as OPTIML R6. Dataset and Methodology The analysis uses 2,865 observations containing indoor temperature, indoor relative humidity, outdoor temperature, and outdoor relative humidity measurements. Following feature and target construction, 2,773 eligible forecast origins were obtained. The models were evaluated using a within-dataset forward comparison involving 558 test origins. Additional nested forward and held-out-dataset evaluations were conducted to examine the consistency of the results. Experimental Evaluation The study evaluates forecasting performance using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), with comparisons against a persistence baseline. It also examines prediction error distributions, extreme prediction cases, threshold-based classification counts, and model interpretability using Kernel SHAP. The repository is intended to support inspection of the implementation, experimental methodology, and reported analyses. Repository Contents The reproducibility package may include: Source code for data preprocessing and feature engineering. Implementation of the OPTIML stacked ensemble framework. Implementation of the exploratory OPTIML R6 residual extension. Model configurations and hyperparameter settings. Forecasting and evaluation scripts. Prediction-level outputs and experimental results. SHAP-based interpretability outputs. Supporting figures, tables, and documentation. Software dependency and execution instructions. Reproducibility The accompanying documentation describes the software environment, data requirements, execution procedure, and experimental workflow. Users should consult the README and dataset documentation before reproducing the analyses. Keywords Medical cold chain; Internet of Things (IoT); temperature forecasting; multi-horizon prediction; stacked ensemble learning; persistence baseline; explainable artificial intelligence; Kernel SHAP; OPTIML.