4.1. Broiler Performance and Environment
In the dataset explored in this study, no effects of the T above the target were observed, except for mortality between 28 and 35 d. These farms are located in the Andean tropical conditions, which are mountains with microclimates where the T at night and early in the morning always reduces below brooding targets. Considering the low housing insulation and open-sided structures, a low T could be a more prevalent problem than heat stress in these mountain farms.
During the first three weeks of age, cumulative exposure to 2.5 °C below the recommended T correlated to poorer BW, BW gain and FCR, and high mortality during the production cycle. In a recent study, Su et al. [
48] demonstrated that chickens exposed to 3 °C lower than the control group from 8 d onwards resulted in similar FCR and BW up to three and six weeks, respectively, compared to chickens kept under thermoneutral conditions. In contrast, chickens exposed to 12 °C lower than the control group from 8 d onwards were significantly lighter and less efficient than their counterparts from two up to six weeks of age.
In another report, Candido et al. [
13] indicated that a mild cold T during the early stages (27 °C first week and 24 °C in the second week) did not affect BW and FCR from 0 to 21 d, while the most significant BW gain was observed in chickens subjected to these Ts. According to those studies, 3 °C below the ideal T was still within the versatile T range for chicks, preventing a reduction in the live performance. Although in the current study, no significant correlations of T lower than 1.5 nor 2.5 °C with live performance parameters were detected within the first week of age, the sensitivity of chickens to lower T seemed to increase with age since more exposure to mild cold T worsened FCR at 14 d and BW and FCR at 21 d. Other authors that have reported similar results [
2,
3,
5,
6] have indicated that an increase in the basal metabolic rate and energy metabolism due to cold stress led to the rise in chicken energy requirements, which explained the diminished BW or increased FCR observed here. Zhou et al. [
6] also suggested that these effects were related to a redistribution of nutrients during the growth toward thermoregulatory responses.
On the other hand, chickens mainly exposed to intermittently T below 2.5 °C from recommendations up to four weeks of age were correlated with high mortality and worsened FCR at 28 and 35 d. Similarly, Bruzual et al. [
49] demonstrated that chicks raised to 12 d under cooler brooding T (26 ± 4 °C first week and 24 ± 4 °C second week) presented 1.71% more mortality compared to chicks subject to warmer T (32 ± 4 °C first weel and 29 ± 4 °C second week). Another report indicated that the highest mortality at 42 d resulted from chicks exposed to 26.7 °C in the first week, compared to other treatments (4.79 vs. 2.36%) that ranged between 29.4 and 35 °C [
50]. The information obtained from the analyzed dataset suggested that a low T in commercial operations might lead to mortality at the end of the production cycle. It is also necessary to consider that previous reports from the literature described in the present study have been conducted under experimental conditions, achieving desired environmental T and RH through environmentally controlled chambers, which significantly differed from environmental conditions under commercial operations that mainly reflect variability in house environmental conditions.
Additionally, environmental conditions in the tropics may remarkably influence the poultry house environment due to the open-sided structure, representing 90.14% of all houses evaluated in this company. Thus, cycling T between day and night or hourly, precipitations, region, geographic altitude, cold wind drafts from mountains, ventilation and heating systems, and other variables could drive the environmental response observed in this poultry company. It is assumed that mainly the highly variable indoor environmental conditions (
Figure 4) resulted in more cumulative hours of exposure to the house T and RH below recommendations. These house variable conditions affected broiler flocks during the first two weeks since it is more difficult for chickens to adapt to variable environments [
51], and consequently, live performance at slaughter age was negatively affected.
On fitting the RH data ranged from 30.32 to 92%, only a few mild correlations were observed up to 28 d, while linear models among correlated variables had low R
2 (0.17–0.26). Similarly, Zhou et al. [
21] reported that average daily feed intake, FCR, and mortality were not affected when chickens were exposed to RH of 35, 60, or 85%. In contrast, Weaver and Meijerhof [
19] detected broilers to be on average 32 g lighter at 42 d when they were either subjected to weekly increments of 8% of RH from 40 to 80% or raised at a constant 75% RH. Yahav [
20] indicated that BW at 35 d was reduced when chickens were reared in environments with RH less or greater than 60–65% compared to other treatments (40–45, 50–55, and 70–75%), while no differences among treatments were observed on FCR or BW at 28 d. However, it was described [
20] that the response varied at different ambient Ts (28 or 30 °C). Then, RH could rely on an ambient T to affect live performance. At 28 °C, Yahav [
20] determined that the heaviest broilers were observed between 60 and 65% of RH compared to other treatments. Chickens within RH treatments of 50–55% and 70–75% but at 30 °C improved their BW (3.74%), while the 40–45% RH treatment obtained the lightest chickens in both scenarios at 28 (1438 g) and 30 °C (1398 g). Zhou et al. [
21] also indicated that although average daily feed intake was not affected by RH treatments at 26 °C, this parameter was reduced by 35 and 85% RH at 31 °C.
THI analyses indicated that as hours of exposure from 14 and 21 d under THI levels below the combined Aviagen recommended T and RH increased, the FCR at 21 d worsened by 0.007 g/g. In contrast, Purswell et al. [
27] did not detect effects on live performance parameters between 49 and 63 d when chickens were exposed to a THI between 14.8 and 20.7 °C. The BW, BW gain, feed intake, and FCR diminished in that study with higher THI levels. It is assumed that the well-developed feather cover that those animals could have at 49 d helped to resist lower THI levels and resulted in similar responses at slaughter age. Contrarily, chickens at 21 d from the current study may not have a good development of feathers, which causes exposure to slightly lower THI to increase mortality. According to the results presented herein, BW, BW gain, and FCR of chickens were more affected by a low T during the first three weeks of age. At the same time, mortality was associated with an increase of up to 125 h between 0 and 28 d in the exposure to a T 2.5 °C lower than recommended. The RH between 30.32 and 92% seemed to be a parameter that did not significantly affect the live performance of chickens possible to detect in this data analysis. The THI employed here [
38] did not show a significant relationship with performance variables.
Finally, the methodology described herein based on hours of exposure to different environmental conditions can improve the understanding of the effects of T and RH since mean environmental values, maximum and minimum T per day, could not reflect the actual fluctuation within the poultry house. In conclusion, a lower T than recommended affected the live performance of broilers during the whole production cycle. On the contrary, only a few correlations were observed with a T above the target and RH, while the THI did not depict the effects observed with a lower T.
4.2. Farm Management and Infrastructure Factors and Performance
On farm-associated factors, evaluating 86 farms and 2700 flocks, chickens that were raised at the highest altitude (1670 ± 434 m.a.s.l.) were the heaviest (1836 g) and the most efficient (1.466 g/g) in both males and females compared to those chickens raised at a medium elevation (1446 ± 498 m.a.s.l.). Rachmawati et al. [
29] indicated that the heaviest chickens at 35 d were observed when farms were located above 700 m.a.s.l. In comparison, no differences were detected between chickens raised either in lowlands (<400 m) or middle lands (400–700 m). In that study, the authors suggested that a higher BW could result from a greater feed intake associated with an increment in the maintenance requirements due to a cooler ambient T [
29]. In contrast, it is assumed that higher altitudes may represent lower environmental Ts, reducing the effects of heat stress observed in low-altitude regions.
In the present study, the rice hulls as litter type also worsened the FCR of both males and females compared to wood shaving. The litter material like wood shavings, sawdust, sand, rice hulls, wheat straw, and grass in which chickens are raised is a factor that has been widely evaluated in poultry [
32,
52,
53,
54,
55]. Garcês et al. [
32] reported that chickens raised to 35 d in rice hulls did not significantly differ in BW, FI, or FCR compared to those raised in wood shavings. Nevertheless, the survival rate of chickens reared in rice hulls was 4.8% lower than their counterparts reared in wood shavings [
32]. Still, the number of replicates in that study was limited to only three pens per treatment, which could not be enough to reproduce the results, while the current data analysis utilized, on average, 601 flocks for each litter type.
Similarly, using two replicates per treatment, Ramadan and El-Khloya [
56] showed that live performance parameters or carcass traits were not affected by the litter type when using five different types of litter, including wood shavings and rice hulls. Toledo et al. [
30] conducted a comprehensive meta-analysis indicating that broilers raised in wood shavings presented higher BW and better FCR than those reared on rice hulls or alternative litter, respectively. It has been suggested that rice hulls account for the more significant proliferation of fungi and, consequently, mycotoxins [
57], worsening live performance in broilers.
In addition, the results presented herein indicated that BW in males and females improved when the litter was recycled more than once. Abougabal [
58] observed no significant differences in the BW from 1 to 6 wk of age nor in the FCR at the end of the experiment when evaluating four litter treatments (new litter, 50% new litter–50% reused litter, 100% reused litter, and 100% reused treated litter). Still, the author suggested that recycled litter was not hazardous for broiler production performance. In contrast, an experiment conducted by Garcés-Gudino et al. [
59] in tropical conditions indicated that BW (1922 vs. 1753 g), BW gain (53.7 vs. 48.8 g/d), FCR (1.588 vs. 1.633 g/g), and mortality (1.61 vs. 3.2%) at 35 d were improved when the same litter was used for two or three production cycles. It has been mentioned that the recycling process may have beneficial impacts on the performance of broilers since the litter bacterial environment could work as a probiotic or a direct-fed microbial supplementation in chickens that contributes to improving the immunity and the response against coccidia challenges, which usually affect the live performance of broilers [
59,
60,
61,
62].
The distance from the hatchery to the farm measured in both kilometers and time spent to reach the farm showed that chicks traveling the longest distances were heavier at 35 d. In contrast, Bergoug et al. [
28] indicated that chicks traveling 4 and 10 h had less BW from the placement up to 21 d, where the effect of transportation time disappeared, compared to chicks that traveled less than 5 min to the farm (0 h, control). However, it was further investigated with hatchery shipping reports that the results presented herein could vary because chicks intended to travel long distances are usually scheduled to start the journey at night, which could reduce stress generated by heat, dehydration, and feed and water deprival, in comparison to chicks delivered during the day. Those aspects related to stress during transportation could be more detrimental to chick development and health than the duration of the trip per se [
63,
64,
65,
66]. Still, males and females were more efficient when post-hatch transportation was the shortest.
When evaluating these factors together with the growth parameters, the variable importance analyses from the supervised ML models demonstrated that the BW and FCR at 35 d are highly dependent on the live performance that chickens may exhibit during the second and third week of age. Other parameters like sex, the distance between the hatchery and the farm, and farm elevation were important to predict BW.
4.3. Prediction of Performance with ML
The MLR presented the lowest predictability for BW and FCR at 35 d. In contrast, the RF and ANN had the best model fit to predict both the BW and FCR at 35 d, with RF being the method with the greatest predictability capacity. It suggests that both RF and ANN could be used to predict performance results in broiler operations. Nevertheless, it is important to consider the pros and cons of each ML technique before scaling to commercial settings. Some studies have demonstrated the use of ANN to predict responses in broiler performance [
67], behavior [
68,
69], and health [
70,
71]. Although those studies exhibited excellent model predictability performance (R
2 > 0.99), those studies were conducted using small datasets from experimental units, including a few individuals, and under controlled conditions with specific treatments and pre-established data structures. Those characteristics do not represent the normal variability observed in data from commercial broiler operations. A few studies have explored using RF to predict responses in poultry, mainly for broiler breeders under precision feeding [
72,
73,
74] or using sound data [
75]. Only a few reports have used broiler commercial data to predict relations among the variables [
34,
76], indicating that RF was also the best predictive methodology.
The variation of the FCR was explained in less than 50% with these ML techniques under five-fold cross-validation. Based on the present data, it was concluded that the contribution of management, environment, and infrastructure to the FCR at 35 d is less than 50%. Most likely, its variability is associated with other factors like nutrient intake, which was not included in those models. Finally, BW at 21 d, sex, distance from hatchery to farm, feed intake, and farm altitude were the five most important variables to predict the BW at 35 d. In comparison, FCR at 21 d, sex, farm altitude, feed intake at 21, and FCR at 14 d accounted for the most essential factors in predicting FCR at 35.