Exploring machine learning techniques for symptom-based detection of livestock diseases

Abstract

Effective livestock monitoring ensures food security and sustainability in our rapidly growing world. However, proper cattle disease is still not taken seriously in our country. Even in the livestock industry, it has not become important yet. Very few livestock farms in Bangladesh collect data on their cattle, so gaining enough data is very tough. Most farm owners are not interested in collecting data; they fear the cost of IoT-based digital farms. Cost is a major concern for small farms as well. The proposed research aims to analyse the application of ML models in this specific sector of livestock management which is disease detection, by analysing various symptoms. Traditionally, Bangladeshi farms provide initial treatment to cattle based on symptoms. Most veterinary doctors in the village used these techniques as a tool for disease detection. We have worked with a dataset of about 43800 instances where almost 28 symptoms were used to detect a disease accurately. Advanced machine learning models such as Neural Network, Gradient boosting classifier, Decision tree classifier, Random forest, XGBoost, KNN etc. were used to determine possible diseases based on the collected symptoms. Overall, this research seeks to provide valuable insights and proper mitigation techniques into the livestock industry by analysing the impact of disease, as this will reduce mortality rates, fulfil the market demand for protein, and bring benefits to the dairy industry.