







































 

 

 

Vol.6, No.2, July 2025 | 92 

 

P-ISSN: 2715-2448 | E-ISSN: 2715-7199 
Vol.6 No.2 July 2025 
Buana Information Technology and Computer Sciences (BIT and CS) 

Detection of Hijaiyah Letters Handwritten in Early Childhood 

Using Yolo V8 
 

Hidayatul Mustagfiroh1, Aviv Yuniar Rahman2, Rangga Pahlevi Putra3 
1,2,3 Department of Informatic Engineering, Universitas Widya Gama, Malang, Indonesia 

E-mail: hidayatulm356@gmail.com1, aviv@widyagama.ac.id2, rangga@widyagama.ac.id3 

 
Received: 2025/05/27 | Revised: 2025/06/26 | Accepted: 2025/07/28 

 

Abstract  

 

This study investigates the effectiveness of the YOLOv8 (You Only Look Once version 8) algorithm in 

detecting handwritten Hijaiyah letters among early childhood learners. The introduction of technology 

in early childhood education is essential for enhancing literacy skills, particularly in learning the 

Arabic alphabet, which is crucial for reading the Quran. This research addresses the challenges faced 

by educators in assessing children's handwriting, which often lacks consistency and objectivity. A 

dataset of 3,780 images of handwritten Hijaiyah letters was collected from children at RA BAIPAS 

Roudlotul Jannah, including various writing styles to ensure the model's robustness. Prior to training, 

the images underwent preprocessing steps such as resizing, normalization, and data augmentation 

techniques like rotation and flipping to enhance the quality and diversity of the training data. The 

YOLOv8 model was trained using an 80-10-10 split for training, validation, and testing datasets. 

Evaluation metrics such as precision, recall, and mean Average Precision (mAP) were used. The results 

showed that YOLOv8 achieved an impressive accuracy of 96.08% in detecting handwritten Hijaiyah 

letters, with high precision and recall rates further validating the model's reliability. This research 

highlights the potential of integrating advanced object detection algorithms like YOLOv8 into 

educational practices. By providing real-time feedback, the system can significantly enhance the 

learning experience for young children, facilitating their understanding and mastery of the Arabic 

alphabet. Future research should focus on expanding the dataset and refining the model to address 

handwriting variability challenges and improve accuracy. 

 

Keywords: Arabic alphabet, early childhood education, handwriting recognition, Hijaiyah letters, 

object detection, and YOLO. 

 

 

I. Introduction 

The hijaiyah letter is a letter to arrange 28 words in Arabic with different forms [1]. However, there 

are other sources that mention it with other numbers. Among them, there are 28 and 30 [2]. Including 

Lamalif, which some people consider to be a different letter [3]. So that without Lamalif the total 

number is 29 hijaiyah letters. In addition, the generally accepted count, excludes Lamalif and combines 

Alif with Hamzah [4]. So some sources dispute this calculation, emphasizing the complexity of the 

Arabic script [5] . The letter begins with Alif and ends with Ya. Hijaiyah letters are also an integral part 

of the Quran, both as a basis for reading it and understanding its contents [6]. Therefore, learning, 

memorizing, and understanding hijaiyah letters is the first stage to be able to read and understand the 

Quran. 

However, not everyone can immediately know and understand the basic science of the Quran, 

namely hijaiyah letters, especially Early Childhood in kindergarten, namely RA BAIPAS Roudlotul 

Jannah. They generally do not know the shape, how to read, and how to write all the hijaiyah letters 

properly and correctly. Understanding the cognitive development stage of children is very important to 

adjust the teaching method. The learning process involves recognizing letter shapes, understanding 

simple words, and reading the signs of the letters [7]. 



Vol.6, No.2, July 2025 | 93 

 

Then there are obstacles when what is assessed is the results of children's handwriting work, for 

example on the object of handwriting hijaiyah letters of the work / work of early childhood. Subjectively 

humans are able to provide an assessment of the work but there are times when it is less consistent and 

difficult to determine with certainty the level of similarity of hijaiyah letter handwriting to the hijaiyah 

letters used as a reference [4]. However, while automated systems can improve accuracy, they may not 

fully capture the nuances of individual handwriting styles, which remain important in evaluating 

children's learning progress. 

One solution to maximize the learning process and useful for teachers to make it easier to correct 

the handwriting of the hijaiyah letters of their students is to use object detection [8]. This research tries 

to develop object detection of hijaiyah letter writing in early childhood, one of which is the application 

of object detection, especially using the YOLO (You Only Look Once) method, can significantly 

improve the learning process to recognize hijaiyah letters in early childhood education [9],[15]. The 

detection system using YOLO is proven to be faster and more accurate to detect an object in an image 

or image so that it is most suitable if applied to the case taken by the researcher [10]. Using object 

detection can help teachers in recognizing and distinguishing each of the 30 hijaiyah letters that a person 

will learn. 

II. Methods 

This chapter will discuss the methods used in the process of detecting children's handwritten 

hijaiyah letters. This research method is a method used to detect early childhood handwritten hijaiyah 

letter objects using YOLOV8. This research uses a quantitative approach with an experimental design 

that aims to test the effectiveness of the Hijaiyah letter detection system. In this research design stage, 

it is presented in the flowchart illustration in Figure 1. The steps are carried out sequentially in order to 

get maximum results in writing the final report. 

 

 
. Figure 1. Research design. 

(Source: Personal Preparation) 



 

Vol.6, No.2, July 2025 | 94 

 

The image above illustrates the step-by-step process involved in building and training a model to 

recognize handwritten Hijaiyah letters. 

 

1. Dataset Retrieval 

The first step is the retrieval of datasets, which involves taking photographs of early childhood 

handwritten Hijaiyah letters using a printer scanner. The dataset, consisting of 3,780 images, is 

categorized into 30 classes, with each class containing 126 images of different Hijaiyah letters. 

These images serve as the training samples for the system being built. 

2. Data Preprocessing 

Once the dataset is collected, it goes through a preprocessing stage. During this stage, the images 

are resized to ensure consistency in dimensions across the entire dataset. This step is critical for 

optimizing the images and ensuring that they meet the input requirements of the model being 

trained. Consistent image size allows the model to process and learn from the data efficiently. 

3. Labeling Dataset 

The next step is labeling the dataset, which involves marking each image with a bounding box 

using labeling software. This step is essential for training the model to recognize the objects (in 

this case, Hijaiyah letters) within the images. Each bounding box represents the object in the image, 

allowing the system to identify and classify it accurately. 

4. Split Dataset 

After labeling the dataset, the data is divided into three parts: training data, validation data, and 

testing data. The training data is used to train the model, while the validation data is utilized to 

evaluate the model's performance during training. The testing data, which has never been seen by 

the model before, is used to assess the model's ability to generalize and perform accurately on new, 

unseen data. 

5. Training 

At this stage, the divided data—training and testing—are accessed through an API in Google 

Colab. The model is trained using the YOLOv8 algorithm, with accuracy being the primary 

parameter for evaluating performance. If the initial test results show low accuracy, retraining is 

performed. This retraining aims to optimize the model to improve its accuracy and overall 

prediction capability. 

6. Detection Result 

The detection process begins by feeding the collected images into the system, which uses the 

YOLOv8 algorithm to analyze the images and recognize objects. The model then marks the 

detected objects with bounding boxes, providing a clear distinction between the objects and the 

background. The output of this process is an accuracy score, which reflects how well the model is 

able to identify the objects within the images. 

7. Model Evaluation 

Finally, the model's performance is evaluated based on its ability to recognize Hijaiyah letters in 

previously unseen images. Evaluation metrics include the detection accuracy, inference time, and 

other relevant parameters that measure the effectiveness of the model. This evaluation ensures that 

the model performs well in recognizing objects under real-world conditions. 

  



Vol.6, No.2, July 2025 | 95 

 

 

III. Results and Discussions 

The results of this study explain the object detection of early childhood handwritten hijaiyah letters 

using YOLOV8. With this discussion, it can be seen the success in detecting children's handwriting 

objects. Datasets that have been labeled will be resized and divided into 3 types of data, namely 80% 

train data, 10% valid data and 10% test data. to facilitate the author in dividing label data quickly 

without having to sort out one by one [12]. The dataset sharing process is shown in Figure 2. 

 

 
Figure 2. Split dataset Anaconda 

(Source: Personal Preparation) 

 

After the installation of the YOLOv8 algorithm system, the training process with the YOLOv8 

algorithm uses the model training configuration, namely images with a width resolution of 640 with 

adjusting height, a total of 50 epochs and 10 batches. The model configuration process can be seen in 

Figure 3. 

 
Figure 3. Training model YOLOv8 

(Source: Personal Preparation) 

 

The results of the evaluation of the YOLOv8 model on the validation set consisting of 378 images 

showed excellent performance, with an accuracy of 0.973 and a recall of 0.5597. mAP50 reached 

0.6129, and mAP 50-95 of 0.4147. The average processing time per image is 4.5 ms for inference, 

which indicates its reliable real-time detection capabilities. The results of the evaluation of the YOLOv8 

model can be seen in Figure 4. 

 
Figure 4. YOLOv8 Evaluation Results 

(Source: Personal Preparation) 

 

After the model will be evaluated using test data to measure its overall performance. This involves 

using evaluation metrics such as Precision, Recall, and mAP to measure how well the model can detect 

hijaiyah datasets in the image [14]. results as shown in Table 1. 

 



 

Vol.6, No.2, July 2025 | 96 

 

Table 1. Results of the evaluation of the YOLOv8 Model (Source: Personal Preparation) 

Class Name Box(P) Recall 
mAP 

50 

mAP50-

95 

Class 

Name 
Box(P) Recall 

mAP 

50 
mAP50-95 

alif 0,947 0,949 0,961 0,766 tha 0,972 1 0,995 0,867 

ba' 0,956 0,789 0,943 0,79 dha 0,973 1 0,995 0,783 

ta 0,925 1 0,938 0,723 ain 0,997 1 0,995 0,686 

tsa 0,92 0,828 0,862 0,622 ghain 0,981 0,938 0,946 0,698 

jim 0,984 1 0,995 0,698 fa 0,996 1 0,995 0,778 

kha 0,777 1 0,871 0,685 qaf 0,979 1 0,979 0,649 

kho 0,99 1 0,995 0,773 kaf 0,92 1 0,979 0,641 

dal 0,848 1 0,942 0,705 lam 0,937 0,931 0,934 0,602 

dzal 0,764 0,778 0,827 0,616 mim 0,986 1 0,968 0,687 

ra 0,981 1 0,938 0,683 nun 0,925 0,867 0,914 0,674 

zai 0,939 0,853 0,949 0,742 waw 0,862 0,965 0,968 0,617 

sin 0,993 1 0,995 0,78 ha 1 0,964 0,995 0,761 

syin 0,981 1 0,995 0,731 lam alif 0,821 1 0,898 0,561 

shad 0,974 1 0,995 0,7 hamzah 1 0,925 0,995 0,75 

dhad 0,958 0,923 0,95 0,669 ya 0,922 0,984 0,983 0,786 

 

The results of the evaluation of the YOLOv8 model on the validation set consisting of 378 images 

showed excellent performance, with an accuracy of 0.973 and a recall of 0.5597. mAP50 reached 

0.6129, and mAP 50-95 reached 0.4147. The average processing time per image is 4.5 ms for inference, 

which shows reliable real-time detection capabilities. 

 
Figure 5. Precision-Recall Curve 

(Source: Personal Preparation) 

 

From Figure 5. It can be seen that the level of accuracy is quite good, even until someone touches 

the number 16 which means very accurate, for example the letter Ghain. The detection of hijaiyah letters 

using YOLOv8 went well and the accuracy value was quite high. In Table 2. Explain the results of the 

hijaiyah letter detection test 



Vol.6, No.2, July 2025 | 97 

 

 

 

Table 2. Test Results (Source: Personal Preparation) 

Test Data Dataset Class Detection Results Level of Certainty 

 

Shad 

 

1.0 

 

Tsa 

 

0.8 

 

Dzal 

 

0.8 

 

Alif 

 

1.0 

 

Ba 

 

0.9 

 

IV. Conclusions 

The application process for the detection of hijaiyah letters written by early childhood used 

YOLOv8 with a lancer and succeeded in accurately detecting the presence of breeders. The dataset used 

consisted of 3790 images on hijaiyah letters which were divided into 30 classes. This dataset is divided 

into 3 parts: training data (80%), validation (10%), and testing (10%) [12].  

Detection using the YOLO model yielded an accurate accuracy of 0.9608 drawn from the mAP50 

results, 0.973 precision, and 0.98 recalls.  The graph shows a precision average value (mAP) of 0.962 

for all classes. This presentation shows that the real-time object detection system using YOLOv8 

provides accurate and reliable results when tested.  

  



 

Vol.6, No.2, July 2025 | 98 

 

References 

 

[1] Y. Mohamed, S. Ismail, and Y. Suryadama, “Pronunciation of Hijaiyyah’s letter for New 

Quranic Learners a Contrastive Analysis Study,” Ulum Islam., vol. 36, no. 01, pp. 73–82, 2024, 

doi: 10.33102/uij.vol36no01.553. 

[2] Asmaa Rafat Elsaied, “Relationship between Numbers and Letters,” J. Math. Syst. Sci., vol. 6, 

no. 8, pp. 335–337, 2016, doi: 10.17265/2159-5291/2016.08.005. 

[3] L. Sarifah, S. Khotijah, and M. K. Khaliqah, “Identification Of Hijaiyah Letters Image Using 

Extreme Learning Machine Method,” J. Mat. Stat. dan Komputasi, vol. 20, no. 1, pp. 90–101, 

2023, doi: 10.20956/j.v20i1.27158. 

[4] M. M. Bahjat, E. Sayed, M. Salem, and A. A. Ghafoor, “A Lexicon of Basic Vocabulary in the 

Holy Quran: The ‘Hamza’ Character as a Model,” vol. 6, no. 1, pp. 56–67, 2023, doi: 10.18860 

/ijazarabi.v6i1.17642. 

[5] A. Saber, A. Taha, and K. Abd El Salam, “A Comprehensive Approach to Arabic Handwriting 

Recognition: Deep Convolutional Networks and Bidirectional Recurrent Models for Arabic 

Scripts,” Int. J. Telecommun., vol. 04, no. 02, pp. 1–11, 2024, doi: 

10.21608/ijt.2024.291347.1052. 

[6] R. F. Rahmat, F. Akbar, M. F. Syahputra, M. A. Budiman, and A. Hizriadi, “An Interactive 

Augmented Reality Implementation of Hijaiyah Alphabet for Children Education,” J. Phys. 

Conf. Ser., vol. 978, no. 1, 2018, doi: 10.1088/1742-6596/978/1/012102. 

[7] K. Stein-smith et al., “The Independent Self-Directed Language Learner and the Role of the 

Language Educator — Expanding Access and Opportunity Kathleen,” J. Lang. Teach. Res., vol. 

14, no. 2, pp. 5–13, 2023, doi: https://doi.org/10.17507/jltr.1402.01. 

[8] Siti Mahrami Ivlatia, Nina Wandana, Dita Andini Harahap, Aslam Annashir, and Sahkholid 

Nasution, “Analisis Kompetensi Penulisan Huruf Hijāiyah Tunggal Pada Siswa MIS UMMI 

Lubuk Pakam,” Semant.  J. Ris. Ilmu Pendidikan, Bhs. dan Budaya, vol. 2, no. 1, pp. 188–200, 

Jan. 2024, doi: 10.61132/semantik.v2i1.284. 

[9] H. Sidi, A. Yuniar, and F. Marisa, “Expression Detection of Children with Special Needs Using 

Yolov4-Tiny,” vol. 16, no. 3, pp. 221–227, 2025. 

[10] A. Y. Rahman and Z. Zakaria, “Hybrid YOLOv8 and Fast R-CNN for Accurate Schematic 

Detection in Power Distribution Networks,” IEEE Access, vol. PP, p. 1, 2025, doi: 

10.1109/ACCESS.2025.3561279. 

[11] J. J. P. Jansen, C. Heavey, T. J. M. Mom, Z. Simsek, and S. A. Zahra, “Scaling-up: Building, 

Leading and Sustaining Rapid Growth Over Time,” J. Manag. Stud., vol. 60, no. 3, pp. 581–

604, May 2023, doi: 10.1111/joms.12910. 

[12] V. C. Raykar and A. Saha, “Data split strategies for evolving predictive models,” Lect. Notes 

Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 9284, 

pp. 3–19, 2015, doi: 10.1007/978-3-319-23528-8_1. 

[13] Z. Q. Zhao, P. Zheng, S. T. Xu, and X. Wu, “Object Detection with Deep Learning: A Review,” 

IEEE Trans. Neural Networks Learn. Syst., vol. 30, no. 11, pp. 3212–3232, 2019, doi: 

10.1109/TNNLS.2018.2876865. 



Vol.6, No.2, July 2025 | 99 

 

[14] R. Padilla, W. L. Passos, T. L. B. Dias, S. L. Netto, and E. A. B. Da Silva, “A comparative 

analysis of object detection metrics with a companion open-source toolkit,” Electron., vol. 10, 

no. 3, pp. 1–28, Feb. 2021, doi: 10.3390/electronics10030279. 

[15] Priyatna, B., Rahman, T. K. A., Hananto, A. L., Hananto, A., & Rahman, A. Y. (2024). 

MobileNet Backbone Based Approach for Quality Classification of Straw Mushrooms 

(Volvariella volvacea) Using Convolutional Neural Networks (CNN). JOIV: International 

Journal on Informatics Visualization, 8(3-2), 1749-1754. 


