Individual feeding patterns as indicators of pig welfare

Electronic feeders (EF) are used in group-housed sows and fattening pigs. EF record animal identification, feed intake and time stamps of entering and leaving a feeder. Individual feeding patterns (e.g. feeding frequency, rate and time and meal duration) can be extracted from these data, though little is known about the variation across and within days of the basal diurnal patterns. We studied variation in diurnal feeding patterns between pigs and within a pig over time. Data was collected at a farm using 12 EF (Nedap) within one group of ~640 gestating sows, and at a farm using single-space EF (Hokofarm group) in pens with 11 fattening pigs. We included records of 239 sows (parity 1-9) with 100 days of feeding data of one gestation and of 110 fattening pigs during their growing period of 95 days. In sows, data was cleaned and aggregated to daily and full gestation period and in fattening pigs to the hourly level and scaled at pig level by dividing hourly by daily intake. Sows visited EF on average 3±2.8 times/day, but mostly ate their allowance at once in 18±3.7 min during the first visit with a feeding rate of 151±18.7 g/min. All feeding patterns showed differences in distributions between sows and effects were found for the hour of first visits, rank in feeding order, number of visits without feed allowance and feeding rate (all P<0.01; Kruskal-Wallis). In fattening pigs, self-organising maps (dissimilarity based on weighted cross-correlation, WCC) combined with hierarchical clustering showed that diurnal patterns could be summarised in 8 clusters, differing in the number, height, width and timing of intake peaks. WCC-coefficients at lag=1d were between 0.40-0.80 (median 0.61) for individual pigs, suggesting quite some pigs show similar patterns in adjacent days. As the lag increased, WCC-coefficients decreased (lag=28d, median 0.58, range 0.28-0.75), reflecting a slow change in diurnal patterns over time. We conclude that pigs show individual feeding patterns, which develop over time. This can help to identify normal and deviating patterns and improve detection of welfare problems via EF.