# TBESO-BP: an improved regression model for predicting subclinical mastitis

## TL;DR

This paper introduces TBESO-BP, an improved neural network model that accurately predicts subclinical mastitis in dairy cows using DHI data.

## Contribution

The novel TBESO algorithm enhances BP neural networks for better prediction accuracy and efficiency in subclinical mastitis detection.

## Key findings

- TBESO-BP achieved an R2 of 0.94, MAE of 2.07, and RMSE of 5.33 in predicting subclinical mastitis.
- TBESO-BP outperformed six alternative models in accuracy and error reduction.
- The TBESO algorithm improves BP networks by addressing issues like local optimization and population diversity.

## Abstract

Subclinical mastitis in dairy cows carries substantial economic, animal welfare, and biosecurity implications. The identification of subclinical forms of the disease is routinely performed through the measurement of somatic cell count (SCC) and microbiological tests. However, their accurate identification can be challenging, thereby limiting the opportunities for early interventions. In this study, an enhanced neural backpropagation (BP) network model for predicting somatic cell count is introduced. The model is based on TBESO (Multi-strategy Boosted Snake Optimizer) and utilizes monthly Dairy Herd Improvement (DHI) data to forecast the status of subclinical mastitis in cows.

The Monthly Dairy Herd Improvement (DHI) data spanning from January 2022 to July 2022 (full dataset) was partitioned into both the training and testing datasets. TBESO addresses the challenge associated with erratic initial weights and thresholds in the BP neural network, impacting training outcomes. The algorithm employs three strategies to rectify issues related to insufficient population diversity, susceptibility to local optimization, and reduced accuracy in snake optimization. Additionally, six alternative regression prediction models for subclinical mastitis in dairy cows are developed within this study. The primary objective is to discern models by exhibiting higher predictive accuracy and lower error values.

The evaluation of the TBESO-BP model in the test phase reveals a coefficient of determination R2 = 0.94, a Mean Absolute Error (MAE) of 2.07, and a Root Mean Square Error (RMSE) of 5.33. In comparison to six alternative models, the TBESO-BP model demonstrates superior accuracy and lower error values.

The TBESO-BP model emerges as a precise tool for predicting subclinical mastitis in dairy cows. The TBESO algorithm notably enhances the efficacy of the BP neural network in regression prediction, ensuring elevated computational efficiency and practicality post-improvement.