Thrips (Thysanoptera) identification using artificial neural networks.

Area of Science:

• Entomology
• Computational Biology
• Machine Learning

Background:

• Thysanoptera (thrips) identification can be challenging due to morphological similarities.
• Accurate identification is crucial for ecological studies and pest management.
• Semi-automated methods can improve efficiency and consistency in species identification.

Purpose of the Study:

• To evaluate the effectiveness of a supervised artificial neural network (ANN) for semi-automated identification of 18 common European Thysanoptera species.
• To assess the accuracy of the ANN model using morphometric and qualitative characters.
• To determine the potential of ANN for practical application in Thysanoptera identification.

Main Methods:

• A supervised artificial neural network (ANN) model, specifically a multilayer perceptron with a single hidden layer, was developed.
• Input data comprised 17 continuous morphometric and two qualitative characters from various body parts (head, pronotum, forewing, ovipositor) and sex.
• The model was trained and tested on a dataset of 498 thrips specimens.

Main Results:

• The ANN model achieved a high accuracy rate of 97% for simultaneous identification of both males and females across all 18 species.
• The classification was performed on an independent test dataset, validating the model's reliability.
• A simple ANN architecture proved sufficient for robust species identification.

Conclusions:

• Supervised artificial neural networks offer a reliable and efficient tool for semi-automated identification of European Thysanoptera species.
• The high accuracy achieved suggests ANN models can be valuable in entomological research and pest control practices.
• This approach has the potential to significantly aid taxonomists and researchers in routine identification tasks.