Machine Learning Approach to Support Taxonomic Discrimination of Mayflies Species Based on Morphologic Data

Area of Science:

• Entomology
• Taxonomy
• Computational Biology

Background:

• Traditional mayfly identification is time-consuming.
• Artificial intelligence (AI) and machine learning (ML) offer objective solutions for taxonomic keys.
• The genus Americabaetis is diverse in South American freshwater ecosystems.

Purpose of the Study:

• To apply ML for identifying species within the mayfly genus Americabaetis.
• To develop an efficient method for mayfly species identification using morphological data.
• To explore AI's potential in taxonomic research.

Main Methods:

• Utilized ML, specifically decision trees with the Gini algorithm, to analyze morphological data.
• Examined specimens from the Museu de Entomologia da Universidade Federal de Viçosa (UFVB/Brazil) and literature data.
• Analyzed eleven morphological traits, including frontal keel, mouthpart shape, and abdominal color pattern.

Main Results:

• Successfully differentiated eight Americabaetis species using only eight morphological characters.
• The ML model effectively identified 40% of the known species in the genus.
• Identified distinct groups within Americabaetis alphus based on abdominal tracheae pigmentation variations.

Conclusions:

• AI and ML provide a novel, efficient approach to mayfly species identification.
• This study integrates biological collections, literature, and AI for taxonomic advancement.
• The developed tool aids in the identification of contemporary and extinct mayflies.