Method finds missing data in precision livestock systems

Photo: Swedish University of Agricultural Sciences.

With the growing adoption of precision livestock farming practices and artificial intelligence in animal handling systems, a great deal of data is generated, but as is the case in many “big data” applications, there are missing data points that need to be accounted for.

According to Keni Ren, a postdoctorate researcher with the Swedish University of Agricultural Sciences Department of Animal Breeding & Genetics, the field of statistics has a number of interpolation methods that help improve data quality, by predicting and estimating the unknown data. Ren and her colleagues decided to find out which interpolation method would be the most accurate one to use when studying the positioning data for dairy cows inside a free-stall dairy farm.

“We investigated four different types of methods. All four are well known methods but we didn’t know how much they differ or which one to prefer in this specific application, with real cow life,” Ren said.

For her study, 69 cows were observed in detail for six days to analyze where the missing data happened and how long the gaps lasted. The data from the 20 most reliable tags were compared with the interpolations — of simulated missing data — from the four different interpolation methods: (1) previous position, (2) linear interpolation, (3) cubic spline data interpolation and (4) modified Akima interpolation.

“The modified Akima interpolation method showed the lowest error distance for all investigated cow activities; walking, feeding, resting and standing. In other words, it had the highest prediction accuracy for the various activities in the barn,” Ren said.

“The advantage of using the Akima method, compared to the other methods, is the greatest when it comes to filling in the gaps of missing data occurring when the animal is walking. I was surprised by how the error distance was stabilized after just one minute of missing data, using this method,” Ren noted.

According to Ren, researchers often mention the problems with missing data, when they publish their studies on animal behavior.

“The different interpolation methods are frequently used, but I have not seen anyone verifying which method is the best one for studying the social network of dairy cattle before,” she said.

The research was published in Frontiers in Animal Science, which contains details on the interpolation methods used.