Performance of Classification Models in Japanese Quail Egg Sexing Jesusimo L. Dioses, Jr. Graduate School Technological Institute of the Philippines Quezon City, Philippines jdiosesjr@gmail.com Ruji P. Medina Graduate School Technological Institute of the Philippines Quezon City, Philippines ruji.medina@tip.edu.ph Arnel C. Fajardo College of Computing Studies Information and Communication Technology Isabela State University Cauayan Isabela, Philippines acfajardo2011@gmail.com Alexander A. Hernandez Graduate School Technological Institute of the Philippines Manila, Philippines alexander.hernandez@tip.edu.ph Abstract—The method of identifying the gender of an avian egg before hatching, egg sexing, has been one of the interesting fields of research in poultry and egg industries to improve its production with reduced costs. Researchers started to study and suggested various scientific methods to determine the sex of avian eggs like chicken and duck. The study proposed the extraction of seven (7) Japanese quail egg morphology features using image processing techniques and edge detection models. Kernel Naïve Bayes, Logistic Regression, and Quadratic SVM models tested and validated Japanese quail eggs' extracted morphology data to classify their sexes. Confusion matrices were used to determine the male, female and average sex classification accuracy rate of each model. Results show that two (2) morphology features of the Japanese quail egg, such as eccentricity and shape index, can be used as significant factors in classifying its sexes. Gaussian Naïve Bayes model is the best classifier to test and validate the morphology characteristics and data of Japanese quail eggs. It has a classification rate of 85.14% for males, 80.16% for females, and an average of 82.88% for both sexes. Keywords—egg-sexing, edge detection, egg morphology, feature extraction, machine learning, confusion matrix. I. INTRODUCTION Egg sexing is a process of determining the gender of an avian egg before hatching. It has become one of the interesting areas of study in the poultry and egg industry. Farmers rely on fertile eggs' observed shape to determine if it is male (oval- shaped) or female (round shape). Prior studies proposed different scientific methods to determine the avian egg sex like chicken and duck before its hatching period. One of these methods is correlating the physical or morphology measurements of avian eggs as predictors of their sex. In their study [1], morphology measurements can be used to predict White Layer Chicken egg sex. Applied in other species, the eccentricity of the duck's egg was extracted. The result shows that it can be used as a predictor in egg sexing [2]. To enhance extraction, researchers applied image processing techniques to extract morphology features of an egg to avoid manual measurement limitations and human errors. This method is significant in image acquisition, image preprocessing, image enhancement, image segmentation, feature extraction, and image classification. In this process, the output is any of the characteristics and qualities present in an image source [3]. Morphology features extraction is one of the methods in extracting and processing scientific information from image data directly related to the shape of observed features [4]. This process can be used for pattern recognition to acquire essential information, which can be used in a classification technique's accuracy [5] [6]. Using this process, researchers contributed valuable extracted morphology features in different studies like coffee beans [7] [8], rice grains [9], poultry products [10][11], and fruit and plants [12] [13] which serves as significant inputs for accurate classifications, predictions, and valuable data analysis. Edge detection plays a vital role in different image analyses like in the field of medicine [14], fruit classification, and plant leaf diseases [15] [16]. It is a starting point in an object detection process. It focuses on defining and evaluating the whole image based on the edges identified [17]. It requires a method where it detects the area of amplitude differences in the image points like the location of the image where intensity is changing, which are considered to be the boundary, and separate the plane and image from other locations, which is very much essential to identify the object [18][19]. The sequence of activities is required to retrieve information relevant to the image like sharpening, enhancement, and pre-positioning of the objects to highlights image contrast and filter unnecessary properties or image details [20][21]. II. RELATED LITERATURE Studies in egg morphology feature extraction and measurements using image processing techniques and machine learning models were became common and popular methods and tools in classifying and predicting outputs. In their study [22], thirteen (13) morphology features were extracted from an egg image that serves as machine learning models to predict and classify its weight—the estimated weight of eggs using image processing and artificial neural networks [23]. Fertility and infertility of eggs were detected and classified through Image processing and machine learning models from its captured images [24], [25]. Duck egg gender was identified by [2] using features extracted and a machine learning model. Due to its popularity and demand in the market, Japanese quail egg sexing was included in research studies. In the study conducted by [26], results revealed that egg weight and width could be used as sex predictors. Using weight, color, and shape index extracted in 540 fertile eggs samples and analyzed using multivariate logistic regression with an average accuracy rate of 76.7% was obtained [27]. 2021 IEEE 17th International Colloquium on Signal Processing & Its Applications (CSPA), 5 - 6 March 2021, Langkawi, Malaysia 978-0-7381-4397-2/21/$31.00 ©2021 IEEE 29 2021 IEEE 17th International Colloquium on Signal Processing & Its Applications (CSPA) | 978-1-6654-1484-5/21/$31.00 ©2021 IEEE | DOI: 10.1109/CSPA52141.2021.9377275 Authorized licensed use limited to: Mapua University. Downloaded on February 04,2025 at 15:17:44 UTC from IEEE Xplore. Restrictions apply.