We created a synthetic dataset of 10,000 part-month observations based on the demand distributions and part characteristics reported by Ghobbar and Friend~\cite{ref5} in order to facilitate controlled and repeatable evaluation. The dataset was created to represent the diversity of a realistic MRO catalogue, covering 200 different part types over 50 months: \begin{itemize} \item \textbf{Demand patterns}: intermittent (45\%), lumpy (30\%), and smooth (25\%), with coefficients of variation ranging from 0.3 to~2.8. \item \textbf{Part attributes}: Weights ranging from 0.5 to 850 kg; lead times between 2 and 60 days; unit costs between 500 and 250,000; different material compositions. \item \textbf{Sustainability metrics}: manufacturing energy consumption; recyclability rates from 15 to 95\%; carbon footprint per part from 50 to 15,000~kgCO(_2). \item \textbf{Temporal features}: fleet utilization rates, maintenance program alignment, and seasonal indicators. \end{itemize} The first 40 months (80\% of the total) were used for training, and the final 10 months (20\% of the total) were used for testing. Throughout, five-fold time-series cross-validation was used to prevent overfitting. \subsection{Architectural Integrated Framework} In accordance with the data schema suggested by Benjaafar et al.~\cite{ref6}, the proposed framework connects four sequential stages (Figure~\ref{fig: framework}): \begin{enumerate} \item \textbf{Data Acquisition}: collection of operational records—demand history, lead times, unit costs, energy use, carbon emissions, and supplier environmental disclosures. \item \textbf{Feature Engineering}: Converting raw records into sustainability-enriched features, such as composite environmental impact indicators~\cite{ref10,ref9}, lifecycle assessment scores, recyclability indices, and normalized carbon footprint. \item \textbf{Hybrid ML Forecasting}: An XGBoost-LSTM ensemble that implicitly takes environmental priorities into account while predicting demand, based on part-level sustainability scores~\cite{ref12}. \item \textbf{Multi-Objective Optimization and Circular Economy Engine}: an epsilon-constraint solver that creates a Pareto front across environmental impact, cost, and stockout risk objectives. Its outputs are fed into a recommendation engine that categorizes parts into end-of-life pathways~\cite{ref10}. \end{enumerate}

To enable controlled and repeatable evaluation, we developed a synthetic dataset containing 10,000 part-month observations. This dataset draws from the demand distributions and part characteristics documented by Ghobbar and Friend to simulate the variety found in real MRO catalogs, spanning 200 distinct part types across 50 months: • Demand patterns include intermittent (45%), lumpy (30%), and smooth (25%) distributions, with coefficients of variation from 0.3 to 2.8. • Part attributes encompass weights from 0.5 to 850 kg, lead times from 2 to 60 days, unit costs from 500 to 250,000, and various material compositions. • Sustainability metrics feature manufacturing energy consumption, recyclability rates from 15 to 95%, and carbon footprints from 50 to 15,000 kgCO2 per part. • Temporal features incorporate fleet utilization rates, maintenance program alignment, and seasonal indicators. We allocated the first 40 months (80%) for training and reserved the final 10 months (20%) for testing. Five-fold time-series cross-validation was applied throughout to avoid overfitting. The proposed framework follows the data schema recommended by Benjaafar et al. and integrates four sequential stages: 1. Data Acquisition: Gathering operational records including demand history, lead times, unit costs, energy consumption, carbon emissions, and supplier environmental disclosures. 2. Feature Engineering: Transforming raw data into sustainability-enhanced features such as composite environmental impact indicators, lifecycle assessment scores, recyclability indices, and normalized carbon footprints. 3. Hybrid ML Forecasting: An XGBoost-LSTM ensemble that incorporates environmental priorities while predicting demand using part-level sustainability scores. 4. Multi-Objective Optimization and Circular Economy Engine: An epsilon-constraint solver that generates a Pareto front balancing environmental impact, cost, and stockout risk objectives. The results feed into a recommendation engine that assigns parts to appropriate end-of-life pathways.