Identification and Modeling of Significant External Factors Causing Bovine Mastitis in Cows using ML Methods

Abstract

Bovine mastitis is a condition in which mammary gland has inflammation and the major causative agent is bacterial infection. Microbes usually cause mastitis when they enter teat through teat canal. Streptococci, staphylococci, and gram-negative rods cause most infections. It is an important health issue in dairy farms. Cow welfare and longevity are affected, that can lead to economic losses due to reduced production of milk, poor quality of milk and cost of treatment. The enhancement of herd hygiene conditions is one of several options available today to attain and maintain a good udder health status for dairy cows. To increase an animal’s resistance to mastitis, enhanced mastitis detection methods and genetic selection of animals are used. Mastitis is one of the most frequently occurring and costly disease of dairy cows worldwide. US dairy industry alone has estimated annual cost of US 2 billion dollars. Various measures should be implemented to ensure the health and well-being of the animals that are the foundation of dairy industry. Mastitis is classified as clinical and subclinical type depending on the manifestation of the disease. Clinical mastitis occurs as an inflammatory response to infection and there is visibly abnormal milk with chemical and physical changes. The mammary gland also may exhibit change in its morphology. The subclinical form occurs when both milk and udder appear normal without noticeable manifestations of inflammation. Subclinical mastitis is more prevalent than clinical mastitis and causes the greatest overall losses in most dairy herds worldwide. In general, the economic loss is estimated to be approximately 100 Euros per cow.

It is thus imperative to understand the risk factors that have high association with mastitis. This in- formation can then be used to target those high-value factors to deliver the best impact for prevention

strategies. In order to achieve the aforementioned targets a study has been conducted in which 432 cows are randomly selected from 40 farms in the Rawalpindi and Islamabad. Risk factors that could have a significant impact on mastitis development are selected from the literature for a total of 28 categorical factors. Of these, 18 factors are herd-level and 10 are animal-level. The Surf Field Mastitis (SFMT) test is used in order to classify the diseased vs. normal cases. The purpose of this study is identification of significant external factors causing bovine mastitis in cows

considering local data set of 432 instances keeping in view the issue of multi-collinearity. Our major tar- get is the identification of factors that could differentiate between diseased and normal cow with efficient

and acceptable accuracy. Secondly, development of predictive models using different machine learning

xi

methods like K nearest neighbor (KNN), Support Vector Machine (SVM) and Artificial Neural Network

(ANN) considering a binary dependent variable (state of disease either yes or no) and 28 external fac- tors. After short-listing factors on basis of chi-square analysis, the factors based on assessment of clinical

professionals are selected. The combinations include all set of factors, and then reduced subsets of 10 factors based on chi-square analysis. 14 factors based on assessment of clinical professionals and 7 based on union of statistically

selected factors and subject knowledge. SVM performed best based on 28 factors with 70 percent sensi- tivity while on the basis of 10, 14 and 7 factors ANN performed best as compared to the other models

with 65 percent sensitivity with respect to 10 factors and 70 percent sensitivity with respect to 7 and 14 factors. For model validation, 10-fold cross validation scheme is used. Based on the provided details, the study recommends the use of ANN with 7 factors named factors named feed sharing, washing of udder, condition of udder, dipping (pre post teat dipping), last (milking the mastitis cow last), use of hormones and udder hygiene score to predict category of the target class.