Adaptive neuro-fuzzy inference systems for improved mastitis classification and diagnosis

Area of Science:

• Dairy cattle health
• Computational intelligence in agriculture
• Soft computing applications

Background:

• Mastitis significantly impacts milk quality and dairy farm economics.
• Accurate mastitis detection is crucial for effective management and resource allocation.
• Fuzzy logic models offer potential for managing uncertainty in dairy data.

Purpose of the Study:

• To compare the performance of three adaptive neuro-fuzzy inference systems (ANFIS) for mastitis classification in Holstein dairy cattle.
• To evaluate the effectiveness of gradient descent (GD)-ANFIS, particle swarm optimization (PSO)-ANFIS, and genetic algorithm (GA)-ANFIS.
• To assess the impact of feature reduction techniques (Pearson correlation, PCA) and undersampling on model performance.

Main Methods:

• Employed three ANFIS methodologies: GD-ANFIS, PSO-ANFIS, and GA-ANFIS.
• Utilized Pearson correlation and principal component analysis for feature reduction.
• Applied an undersampling algorithm to address class imbalance in the dataset.

Main Results:

• The GD-ANFIS model combined with the Pearson method exhibited superior performance in mastitis classification compared to PSO-ANFIS and GA-ANFIS.
• Key performance metrics including accuracy, precision, recall, and F1-score were used for evaluation.
• While GD-ANFIS showed strong results, definitive model selection was challenging due to the interplay of multiple criteria.

Conclusions:

• The study highlights the potential of ANFIS, particularly GD-ANFIS with Pearson feature reduction, for improving mastitis detection in dairy cattle.
• Findings contribute to soft computing advancements in precision dairy farming and udder health management.
• Provided MATLAB code for reproducibility and potential development of mobile applications in precision dairy production.