Use of feature selection algorithm to determine the most important factors affecting milk fat percentage of Holstein cows

Monitoring the milk components of a herd can help assess the health and nutritional status of lactating cows. Fat is one of the most valuable components of milk. Milk fat (proportion and total amount) is typically the most variable component in milk which is influenced by various physiological and environmental factors. A herd milk fat test below 0.3% of breed average can indicate a problem situation. Therefore, diets that allow for an improvement in milk fat output would potentially be economically advantageous. In order to maintain milk fat, it is necessary to identify, monitor, and measure the most important factors influencing milk fat. Feature selection algorithm is one of data mining methods and to the best of our knowledge no available literature to use this algorithm to refine most contributing factors affecting milk fat in Holstein dairy cows. Objective of this study was to determine the most important factors affecting milk fat of Holstein cows using feature selection algorithm.
In the current study, a total of 2112 raw data obtained from 66 factors (which might potentially able to affect milk fat) was used. Data was evaluated by the three important methods (Best-First, Greedy-Stepwise, and Ranker methods) and six models (CFS-Subset-Eval, Info-Gain-Attribute-Eval, Gain-Ratio-Attribute-Eval, Symmetricer-Attribute-Eval, RelifeF-Attribute-Eval, Principal-Components) of feature selection algorithm. Data was analyzed using the WEKA Software (v. 3.8).
Results indicated that both Best-First and Gready-Stepwise methods of feature selection algorithm are the most appropriate methods to select efficient factors affecting milk fat using Naive Bayes classification with minimized error. Accordingly, blood non-esterified fatty acid, total volatile fatty acids of rumen fluid, ether-extract intake, time spent rumination per kilogram of neutral detergent fiber intake, time spent rumination per kilogram of forage neutral detergent fiber intake, total chewing time per kilogram of neutral detergent fiber intake, body weight, body condition score, back fat thickness, and blood glucose were the most important factors affecting milk fat.
It is possible to monitor and measure of some limited factors selected by feature selection to efficiently maintain milk fat on-farm and save time and cost.