Introduction

Poultry production has long been a cornerstone of agricultural development, providing a critical source of protein and contributing significantly to food security worldwide. Within the sector, the quality of poultry eggs is a key determinant of both consumer satisfaction and the economic viability of poultry farming operations (FAO, 2019). Various factors influence egg quality, including breed, management practices, and nutritional interventions. Understanding the specific contributions of different chicken breeds to egg characteristics is essential for optimizing production and meeting market demands.

This study considered two chicken breeds commonly namely Fayoumi and Sasso. Fayoumi, a breed known for its hardiness and adaptability to harsh environmental conditions, typically produces smaller eggs with higher yolk content, making it well-suited for local consumption in areas where traditional breeds are preferred (El-Safty et al., 2010). On the other hand, Sasso, a dual-purpose breed developed for both meat and egg production, is valued for its larger eggs, which are preferred in commercial markets (Sezer et al., 2012). Comparing these breeds in terms of egg quality traits such as weight, shell strength, and yolk content provides insights into their suitability for different production systems and markets.

Despite the availability of extensive research on individual egg quality traits, there remains a gap in understanding how these traits can be used to classify breeds and predict egg production performance. Machine learning techniques, such as classification tree algorithms (CTA), offer a promising approach for analyzing complex data sets and identifying patterns that can distinguish between breeds based on their egg characteristics (Jankowski et al., 2017). This study applies CTA to compare the egg quality traits of Fayoumi and Sasso chickens, aiming to identify the most significant predictors of breed differentiation and assess the implications for poultry production systems.

In the context of the study, key egg quality traits such as egg weight (EW), albumen weight (AW), shell weight (SW), and yolk weight (YW) were measured to evaluate their influence on breed classification. The study provides valuable insights into how these traits can be used to enhance breed selection, optimize egg production, and improve poultry management practices.

1. Materials and Methods

2.1 Description of the Study Area:

The study was conducted at the poultry research farm of Haramaya University, situated 505 km to the east of Addis Ababa. The site is positioned at an elevation of 1980 meters above sea level, with coordinates of 9⁰ 26'N latitude and 42⁰ 3'E longitude. The average annual maximum and minimum temperatures are recorded at 23.4°C and 8.25°C, respectively, while the region experiences an average annual rainfall of 741.6 mm.

2.2 Chicken Breeds Management:

In this experiment, two exotic chicken breeds—Fayoumi and Sasso—were used. These breeds were raised under consistent housing and feeding conditions to ensure uniformity. Initially, the chickens were housed in brooder facilities equipped with incandescent heating lamps for the first eight weeks. After this period, they were transferred to the grower house for the growth phase and subsequently moved to the layer house, where they were kept under the deep litter system during the laying phase.

Throughout the experiment, the chickens had continuous access to water and were provided with a nutritionally balanced diet tailored to their specific needs. During the first eight weeks, they were fed a standard ration containing 20% crude protein (CP) and 2800 Kcal/kg metabolizable energy (ME). As they transitioned into the growth phase (9 to 20 weeks), the feed was adjusted to 16% CP and 2800 Kcal/kg ME. Finally, during the laying period, the feed was further modified to 16.5% CP and 1750 Kcal/kg ME to meet the specific dietary requirements of the hens.

To ensure the health and well-being of the chickens, they were vaccinated against major viral diseases. Additionally, a veterinary professional closely monitored their health throughout the study to maintain optimal conditions.

2.3 Data Collection and Traits Measured:

A total of 1,096 eggs, comprising 741 from Fayoumi and 355 from Sasso, were used for this study. The recorded traits included albumen height, albumen weight, egg length, egg width, egg weight, shell weight, yolk colour, yolk height, yolk weight, and shell thickness. Egg weight, length, and width were measured on intact eggs using a sensitive weighing scale and a digital calliper, respectively.

To assess internal egg quality, the eggs were carefully cracked onto a flat glass surface. The albumen and yolk were then separated, and their respective weights were recorded after measuring albumen height with a tripod micrometre. Following this, the eggshells were gently washed and left to air dry for 48 hours, after which shell thickness was determined.

2.4 Statistical Data Analysis:

For all statistical analyses in this study, JMP Pro version 18 (2023) was used.

2.4.1 Analysis of Variance:

Egg quality traits (EW, AH, AW, YH, YW, SW, YC, EL, and EWi) were subjected to a one-way analysis of variance using the general linear model procedure of SAS JMP Pro software to determine the effect of breed. Treatment means were separated using student’s t-test at a 95% confidence interval. The linear model employed was:

Y_ij = μ + B_i + ε_ij

The Classification Tree Algorithm (CTA) was employed to investigate the differentiation of chicken breeds based on egg quality traits. The model's design involved calculating significant differences using p-values derived from Pearson's Chi-square test (Ritschard, 2010). To evaluate the model’s classification accuracy, the area under the ROC curve (AUC) was used. The AUC ranges from 0.0 to 1.0, with 1.0 indicating a perfect positive prediction, 0.0 a perfect negative prediction, and 0.5 representing random or poor prediction (Bayram et al., 2015).

The tree-building process continued iteratively until no further splits could improve the model. The maximum tree was achieved once the model reached its optimal complexity. To estimate future prediction errors, a ten-fold cross-validation method was applied, providing a robust assessment of the error rates for each sub-tree (Camdeviren et al., 2005).

2. Results and Discussions

3.1 Analysis of Variance:

Table 1 below presents the independent t-test results that reveal significant differences in egg quality traits between the Fayoumi and Sasso chicken breeds, highlighting the effect of breed on various egg quality traits. The analysis underscores the importance of breed selection in poultry production, particularly concerning egg quality traits, which are crucial for both consumer preference and industrial applications.

3.2 Classification Tree Algorithm (CTA):

The CTA analysis was employed to discriminate between Fayoumi (FM) and Sasso (SS) breeds based on key egg quality traits. The model revealed critical insights into how these factors differentiate between the two breeds, with varying levels of predictive accuracy across different nodes.

At the root node (Node 0), the classification tree began with a total sample of 821 observations. The Fayoumi breed was dominant in this sample, accounting for 67.60% (n = 555) of the cases, while the Sasso breed made up 32.40% (n = 266). The overall G² statistic was 1034.20, indicating significant heterogeneity between the two breeds in the population.

Node 10 - Impact of Egg Weight (EW): The first split occurred at EW ≥ 52.3 g. This variable had a profound effect on breed classification, particularly for Sasso. In Node 1, out of 270 cases, Sasso dominated, representing 96.67% (n = 261) of the cases with EW ≥ 52.3 g, while Fayoumi contributed only 3.33% (n = 9). The G² value of 78.92 further highlighted the strength of this predictor in separating the two breeds. This result is consistent with previous findings, where larger egg sizes were often attributed to heavier, dual-purpose breeds like Sasso (Sezer et al., 2012).

Node 2 - Egg Weight ≥ 54.2 g: In Node 2, egg weight further refined the classification. With EW ≥ 54.2 g, all 250 cases were classified as Sasso, with a perfect prediction rate for Sasso (100%, n = 250). This result demonstrates the robustness of egg weight as a determinant for identifying Sasso, reflecting its larger size relative to Fayoumi. The absence of Fayoumi in this node underscores the breed’s characteristic of producing smaller eggs, as has been observed in similar research (Hegab et al., 2016).

Node 3 - Egg Weight < 54.2 g: For eggs weighing < 54.2 g (Node 3), the predictive power shifted. While Sasso still accounted for 55% (n = 11) of the cases, Fayoumi started to gain representation, comprising 45% (n = 9). The G² value of 27.53 indicated moderate variation, but egg weight remained an influential trait, consistent with its significance in poultry breed differentiation (De Marchi et al., 2013).

Node 4 - Shell Weight (SW): SW emerged as a further discriminating factor, particularly for lower egg weights. In Node 4, SW < 4.13 g perfectly predicted Sasso, with all 7 cases classified as Sasso (100%). This suggests that Sasso eggs, particularly smaller ones, are characterized by relatively lighter shells. However, for SW ≥ 4.13 g (Node 5), Fayoumi dominated with a probability of 68.55% (n = 9). This aligns with earlier studies showing Fayoumi’s tendency to produce eggs with relatively thicker shells, a trait advantageous for survival in more challenging environmental conditions (El-Safty et al., 2010).

Node 6 - Yolk Weight (YW): YW further refined the classification within eggs that had a shell weight of ≥ 4.13 g. For YW < 17.0 g (Node 6), Sasso predominated at 80% (n = 4), whereas Fayoumi was limited to 20% (n = 1). By contrast, when YW ≥ 17.0 g (Node 7), Fayoumi achieved a perfect prediction rate, representing 100% of cases (n = 8). This result emphasizes the higher yolk content in Fayoumi eggs, a characteristic noted in studies of indigenous breeds (Alkan et al., 2013).

Node 8 - Eggs Weighing < 52.3 g: For eggs weighing < 52.3 g (Node 8), Fayoumi had an overwhelming representation, with 99.09% (n = 546) of cases classified as Fayoumi. The G² statistic of 56.98 reflected substantial differentiation at this point, reinforcing the notion that smaller egg weights are strongly associated with Fayoumi.

Node 9 - Albumen Weight (AW): AW further distinguished between the two breeds. For AW ≥ 28.7 g (Node 9), Fayoumi maintained dominance, constituting 93.98% (n = 78) of the cases. The G² value of 37.79 indicated that AW is a strong predictor for Fayoumi, consistent with the breed’s tendency to produce eggs with higher albumen content relative to Sasso (Hegab et al., 2016). For EL ≥ 5.4 cm (Node 10), Fayoumi was again the majority breed (78.95%, n = 15), while for EL < 5.4 cm (Node 11), Fayoumi constituted 98.44% (n = 63).

Figure 1. Classification tree diagram constructed by CTA

3.2.1 Egg Quality Traits Importance:

The column contribution results (Table 2) revealed that egg weight (EW) was the most significant predictor, contributing the largest portion to the model’s predictive power. Other traits, such as albumen weight (AW), shell weight (SW), yolk weight (YW), and egg length (EL), also played minor roles in differentiating between the two breeds, while traits such as albumen height (AH), yolk height (YH), egg width (EWi), shell thickness (ST), and yolk colour (YC) did not contribute to the splits in the classification.

Table 2: Importance of egg quality traits in the differentiation of the chicken breeds

AH = albumen height, AW = albumen weight, EL = egg length, EW = egg weight, EWi = egg width, SW = shell weight, YC = yolk colour, YH = yolk height, YW = yolk weight, and ST = shell thickness.

Egg Weight (EW): as the Dominant Predictor: EW was found to be the dominant trait in distinguishing Fayoumi from Sasso, accounting for 95.03% of the total G² (G² = 949.70). This strong contribution reflects the fact that Sasso breeds generally produce larger eggs than Fayoumi breeds, a finding consistent with previous studies on breed-specific egg characteristics (Hegab et al., 2016). The two splits based on EW thresholds provided a clear differentiation between the breeds, with heavier eggs typically associated with Sasso and lighter eggs linked to Fayoumi. These results align with the genetic makeup of the Sasso breed, which is selected for both meat and egg production, typically yielding larger eggs (Sezer et al., 2012).

Albumen Weight (AW) and Its Role in Differentiation: AW contributed 1.92% (G² = 19.19) to the overall predictive power of the classification tree. Although a minor contributor compared to egg weight, AW still played a notable role, particularly in conjunction with smaller eggs. Fayoumi eggs, known for their relatively high-quality albumen, tend to display this trait more prominently than Sasso eggs. As previous research has noted, AW is often linked to overall egg quality, making it an important factor in discriminating breeds, especially for smaller eggs (El-Safty et al., 2010).

Shell Weight (SW) and Yolk Weight (YW) Contributions: SW and YW both had modest impacts on the classification tree, contributing 1.15% and 1.11% of the total G², respectively. SW’s role was significant in differentiating eggs within a narrow range of EW, as Sasso eggs tend to have lighter shells than Fayoumi eggs, which are known for their durability (De Marchi et al., 2013). YW’s contribution, though small, helped identify heavier yolk eggs as more likely to belong to Fayoumi, reflecting the breed's trait of producing eggs with larger yolks relative to egg size (Hegab et al., 2016).

Minor Contribution of Egg Length (EL): EL contributed 0.79% (G² = 7.93) to the model, providing additional granularity in distinguishing smaller Fayoumi eggs from the slightly longer Sasso eggs. While this contribution was minor, it highlights the importance of considering multiple egg characteristics when attempting to classify breeds. Research has shown that egg dimensions, including length and width, are often correlated with breed-specific genetic traits (Alkan et al., 2013).

Non-Contributing Traits: Several traits, including albumen height (AH), yolk height (YH), egg width (EWi), shell thickness (ST), and yolk colour (YC), did not contribute to the classification tree, as indicated by their G² values of zero. This suggests that these traits do not provide sufficient discriminatory power between Fayoumi and Sasso breeds. The lack of influence from AH and YH is particularly noteworthy, as these traits are often associated with egg quality but may not vary significantly between these specific breeds (El-Safty et al., 2010). Similarly, traits like shell thickness and yolk colour, which are sometimes used to assess egg quality in other studies, did not play a role in this classification model, likely because they do not differ meaningfully between Fayoumi and Sasso.

3.2.2 Terminal Leaf Reports:

A CTA was applied to differentiate between Fayoumi (FM) and Sasso (SS) breeds using egg quality traits. The tree structure and leaf report (Table 3) indicated how specific traits predicted the likelihood of each breed, with the results highlighting key trait thresholds that effectively separated Fayoumi from Sasso.

Table 3: Terminal leaf report with probability and corresponding count values

AH = albumen height, AW = albumen weight, EL = egg length, EW = egg weight, and YW = yolk weight, BV = Bovan brown, FM = Fayoumi.

Node 1 - Egg Weight (EW) as a Primary Predictor: The first major division in the classification occurred at EW ≥ 52.3 g. This split yielded a nearly perfect prediction for Sasso, with a response probability of 0.998 (n = 250) for eggs at or above this threshold. In contrast, Fayoumi was virtually absent in this category, with a probability of only 0.002. These findings are consistent with prior research indicating that Sasso chickens generally produce larger eggs compared to Fayoumi, a characteristic likely due to their larger body size and genetic predisposition (Sezer et al., 2012).

Node 2 - Impact of Shell Weight (SW): For eggs weighing between 52.3 and 54.2 g, SW became the next critical factor. Eggs with SW < 4.13 g were overwhelmingly classified as Sasso, with a probability of 0.925 (n = 7), further emphasizing the association between lighter shells and Sasso. Conversely, Fayoumi showed a minimal presence in this node (probability = 0.075), a finding that aligns with the documented trend that Fayoumi eggs tend to have thicker, more robust shells (El-Safty et al., 2010).

Node 3 - Yolk Weight (YW) in Intermediate Egg Sizes: For eggs in the same weight range (EW between 52.3 and 54.2 g), the algorithm identified YW as another key predictor. Eggs with SW ≥ 4.13 g were further divided by YW. For eggs with YW < 17.0 g, Sasso remained predominant with a probability of 0.732 (n = 4), while Fayoumi only accounted for 26.76% of the cases (n = 1). However, when YW was ≥ 17.0 g, Fayoumi became the dominant breed, with a response probability of 0.956 (n = 8), reflecting its tendency to produce eggs with larger yolk proportions. This shift in predictive power indicates the importance of yolk size as a distinguishing trait between the breeds, as corroborated by similar findings in indigenous chicken populations (Alkan et al., 2013).

Node 4 - Eggs Weighing < 52.3 g: For eggs weighing < 52.3 g, Fayoumi became the dominant breed. In this category, Fayoumi had a high response probability of 0.990 (n = 546), while Sasso was virtually absent (probability = 0.010). This confirms the consistent production of smaller eggs by Fayoumi, a trait commonly observed in other studies of the breed (Hegab et al., 2016).

Node 5 - Albumen Weight (AW) and Egg Length (EL) as Key Discriminators: Within the category of smaller eggs, AW and EL were crucial predictors. For eggs with AW ≥ 28.7 g and EL ≥ 5.4 cm, Fayoumi was the most likely breed, with a response probability of 0.787 (n = 15), while Sasso had a probability of 0.214 (n = 4). These traits, particularly AW, are significant in differentiating breeds, as higher albumen content is often linked to better egg quality, a trait commonly seen in Fayoumi (Hegab et al., 2016). For eggs with EL < 5.4 cm, Fayoumi's probability increased further to 0.981 (n = 63), with Sasso nearly absent (probability = 0.020, n = 1). This suggests that smaller eggs with shorter lengths are almost exclusively produced by Fayoumi.

Node 6 - Albumen Weight (AW) < 28.7 g: For eggs with AW < 28.7 g, Fayoumi reached near-perfect dominance, with a response probability of 0.999 (n = 468), effectively classifying all these eggs as Fayoumi. This final node clearly establishes the strong association between lower AW and Fayoumi. The absence of Sasso in this category (probability = 0.001) further confirms that this breed rarely produces eggs with such characteristics.

3.2.3 Confusion Matrix Analysis:

The results of the CTA are presented through the confusion matrix (Table 4) for both training and validation datasets. The confusion matrix offers insight into the model's classification accuracy and performance for each breed.

Table 4: Confusion matrix for the training and validation data showing count and percentages

Values in before brackets are counts while those in brackets are per cent.

In the training phase, the model exhibited a high classification accuracy for both breeds. The Fayoumi breed was predicted with 99.8% accuracy (554 correctly classified as Fayoumi, 1 misclassified as Sasso). Similarly, the Sasso breed was classified with 98.1% accuracy (261 correctly classified as Sasso, 5 misclassified as Fayoumi). This demonstrates that the model was highly effective in identifying the breeds based on the input predictors.

The overall misclassification rate was extremely low in the training set, with only 6 out of 821 observations misclassified. This strong performance highlights the algorithm's capacity to handle the egg quality traits as distinguishing factors between the two breeds. This result is consistent with studies demonstrating the potential of machine learning techniques to accurately predict breed classifications using phenotypic traits, such as egg quality characteristics (Kheradmand et al., 2014; Jankowski et al., 2017).

In the validation set, which provides a more objective measure of the model’s generalizability, the classification accuracy remained robust but slightly lower than in the training dataset. Fayoumi was predicted with 98.9% accuracy (183 correctly classified, 2 misclassified as Sasso), while Sasso showed a classification accuracy of 95.5% (85 correctly classified, 4 misclassified as Fayoumi). This slight drop in accuracy for Sasso in the validation phase reflects the challenges often encountered when applying machine learning models to new data, suggesting the potential presence of subtle overlaps in egg quality traits between the two breeds (Tůmová & Chrastinová, 2018).

The misclassification rates in the validation dataset were 1.1% for Fayoumi and 4.5% for Sasso. These results indicate a marginal increase in the error rate compared to the training set but are within acceptable limits, maintaining the model’s credibility in distinguishing between the breeds.

The model's strong performance can be attributed to its reliance on specific egg quality traits that are biologically linked to breed-specific characteristics. Traits such as egg weight (EW) and albumen weight (AW), which were significant contributors to the model, are known to vary significantly between breeds, providing critical discriminatory power (Rizzi et al., 2013).

These results align with existing literature, which emphasizes the role of egg quality traits in breed differentiation, particularly in dual-purpose breeds like Fayoumi and Sasso (Mousa-Balabel et al., 2021).

3.2.3 Receiver Operating Characteristic (ROC) Analysis:

The performance of the CTA, as assessed by the Receiver Operating Characteristic (ROC) curves for both training and validation datasets, indicates highly accurate breed discrimination between Fayoumi and Sasso breeds. The ROC curves (Figure 2) are a critical tool for evaluating classification models, with the area under the curve (AUC) offering a reliable measure of the algorithm's ability to differentiate between classes (Bradley, 1997). In this case, the AUC provides an indicator of how well the model predicted the breeds based on the selected egg quality traits.

Figure 2. ROC diagram for the training (left) and validation (right) data constructed by CTA

The area under the ROC curve (AUC) for both Fayoumi and Sasso in the training dataset was an impressive 0.999. This near-perfect score indicates that the model was highly capable of distinguishing between the two breeds with almost no misclassification. An AUC close to 1.0 signifies that the model had an outstanding ability to discriminate between Fayoumi and Sasso breeds based on the egg quality traits provided, making it a highly reliable predictor (Fawcett, 2006).

This result suggests that the selected predictors, such as egg weight (EW) and albumen weight (AW), contributed effectively to differentiating between the two breeds. Egg quality traits are known to exhibit breed-specific variations, which likely enhanced the model’s ability to make accurate predictions (Mousa-Balabel et al., 2021). The extremely high AUC for the training data underscores the strength of the model when working with familiar data.

In the validation dataset, the model's AUC remained highly favourable, with both breeds achieving an AUC of 0.998. Although this represents a slight decrease from the training data, the results still demonstrate that the model performed exceptionally well when applied to new, unseen data. The minimal reduction in AUC suggests that the model generalizes effectively, and while there may be minor overlaps in egg quality traits between Fayoumi and Sasso, the algorithm retains its capacity to differentiate between the breeds with a high degree of accuracy.

The consistency between the training and validation AUC results (0.999 and 0.998, respectively) highlights the robustness of the classification tree model. Such stability between the two datasets points to minimal overfitting, which is a common concern in machine learning models. Overfitting occurs when a model performs exceptionally well on training data but struggles to generalize to new data, a problem that does not seem to affect this model to any significant degree (Hastie, Tibshirani, & Friedman, 2009).

3. Conclusions

This study successfully compared the egg quality traits of Fayoumi and Sasso breeds, revealing significant differences that underscore the importance of breed selection in poultry production. Sasso eggs were consistently heavier and exhibited superior albumen and yolk quality, making them more suitable for commercial markets. The use of a classification tree algorithm effectively distinguished between the two breeds based on egg weight, shell weight, and yolk weight, demonstrating the potential of machine learning techniques in optimizing breed classification.

While the study provides valuable insights, it is important to acknowledge its limitations, including the focus on only two breeds and a specific geographical context that may limit generalizability. Future research could explore additional breeds and incorporate broader geographic and environmental variables to enhance the applicability of the findings. Further studies may also investigate the impact of dietary and management practices on egg quality traits to provide a more comprehensive understanding of poultry production.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

Financial Support

This research was conducted with the financial support of Haramaya University.