Mel-based Feature Extraction for Acoustic Event Detection and Classification: A Systematic Review and Research Roadmap

• Systematic review of 373 Mel-based AEDC studies from 2020 to 2025. • Five Mel feature types analysed across four key parameters. • Jensen-Shannon divergence quantifies temporal parameter trends. • Domain-specific parameter patterns across four application areas. • Practical guidance for Mel parameter selection in AEDC systems. Acoustic Event Detection and Classification (AEDC) systems are artificial intelligence technologies widely used in modern audio-based monitoring and decision-support applications. These systems may operate independently or function as perceptual modules within intelligent decision-making frameworks and expert systems. AEDC has been applied in diverse domains, including security monitoring, healthcare diagnostics, environmental surveillance, and industrial maintenance. Although Mel-based features are extensively employed in both traditional machine learning and deep learning AEDC approaches, no comprehensive review has systematically analysed their usage patterns and parameter configurations. This systematic review examines 373 AEDC studies published between 2020 and 2025 that employ Mel-based feature extraction. Five principal feature types are analysed: Log Mel Spectrogram (237 studies), Mel Spectrogram (61 studies), Log Mel Band Energies (34 studies), Mel-Frequency Cepstral Coefficients (36 studies), and Mel Band Energies (5 studies). Four critical Mel-related parameters—number of Mel filters, frame length, overlap percentage, and window type—are evaluated to identify dominant configurations, feature-specific patterns, and temporal trends. Classification architectures are systematically mapped across studies to analyse the relationship between feature selection and model design. The results reveal substantial variability in parameter choices across feature types and application contexts, as well as dominant trends aligned with deep learning frameworks. The review further identifies methodological gaps and outlines research directions to improve the design, reporting, and optimisation of Mel-based AEDC systems.