The exponential growth of encrypted and dynamic network traffic poses significant challenges to traditional traffic analysis methods, underscoring the need for robust and scalable solutions. Statistical approaches like L-moments have demonstrated exceptional potential in characterizing traffic flows, offering reduced sensitivity to outliers and the ability to capture higher-order distributional properties with minimal data. Building on previous work by the authors, this study introduces significant enhancements to the L-moment-based methodology for flow analysis and classification, specifically addressing limitations in feature selection and sample size requirements, aspects crucial for achieving deployable configurations in high-performance network environments. Key contributions include the integration of the fifth-order L-moment ratio (𝜏_5 ) for enriched traffic representation and a multi-objective optimization framework based on a multi-objective evolutionary algorithm that balances competing goals: minimizing flow features selected for flow classification, reducing sample sizes for L-moment estimation, and maximizing classification quality. The enhanced methodology was applied to the CIC-DDoS2019 dataset, previously used in the authors’ earlier work, enabling direct comparison. Results show a reduction in sample size requirements from 200 to as few as 10, while simultaneously improving classification accuracy and selecting minimal features. These findings demonstrate the scalability and effectiveness of the proposed framework, designed for resource-constrained environments in Next-Generation Networks (NGNs), and make it publicly available for reproducibility and future research.