IAES International Journal of Artificial Intelligence (IJ-AI) Vol. 11, No. 3, September 2022, pp. 1143∼1152 ISSN: 2252-8938, DOI: 10.11591/ijai.v11i3.pp1143-1152 ❒ 1143 A machine learning approach for Bengali handwritten vowel character recognition Shahrukh Ahsan 1 , Shah Tarik Nawaz 1 , Talha Bin Sarwar 2 , M. Saef Ullah Miah 2 , Abhijit Bhowmik 1 1 Department of Computer Science, American International University-Bangladesh (AIUB), Dhaka, Bangladesh 2 Faculty of Computing, College of Computing and Applied Sciences, Universiti Malaysia Pahang, Pekan, Malaysia Article Info Article history: Received Sep 7, 2021 Revised Jun 17, 2022 Accepted Jun 24, 2022 Keywords: BanglaLekha-isolated Bengali handwritten vowel recognition Handwritten character recognition Machine learning Support vector machine ABSTRACT Recognition of handwritten characters is complex because of the different shapes and numbers of characters. Many handwritten character recognition strategies have been proposed for both English and other major dialects. Bengali is generally considered the fifth most spoken local language in the world. It is the official and most widely spo- ken language of Bangladesh and the second most widely spoken among the 22 posted dialects of India. To improve the recognition of handwritten Bengali characters, we developed a different approach in this study using face mapping. It is quite effective in distinguishing different characters. The real highlight is that the recognition results are more efficient than expected with a simple machine learning technique. The proposed method uses the Python library Scikit-Learn, including NumPy, Pandas, Matplotlib, and support vector machine (SVM) classifier. The proposed model uses a dataset de- rived from the BanglaLekha isolated dataset for the training and testing part. The new approach shows positive results and looks promising. It showed accuracy up to 94% for a particular character and 91% on average for all characters. This is an open access article under the CC BY-SA license. Corresponding Author: Talha Bin Sarwar Faculty of Computing, College of Computing and Applied Sciences, Universiti Malaysia Pahang 26600, Pekan, Malaysia Email: talhasarwar40@gmail.com 1. INTRODUCTION Handwriting recognition has proven to be quite challenging in recent years. Handwritten characters by different people show many complexities as it is not identical and varies in shapes and writing styles [1], [2]. There have been several methods that are introduced for English character recognition. One of the most applicable techniques is by training neural networks for the acknowledgment of characters [3]. At present, Ben- gali is one of the utmost spoken languages, placed around fifth in the world and second among the South Asian Association for Regional Cooperation (SAARC) countries [4]. In almost all phases of life in Bangladesh and in some parts of India, language is used to communicate. Around 220 million individuals worldwide presently uti- lize Bengali to talk and compose reason. A proper machine learning system that works efficiently to recognize its characters is long overdue for such a widely used language. In addition, several works have been done on Bengali character recognition, where it has been challeng- ing to achieve better execution and prediction results due to the natural complexity of most Bengali alphabets. The language has a long and rich scientific heritage of over a thousand years and a history of language evo- lution. Researchers have presented different types of feature extraction techniques and proposed some new feature extraction techniques for recognizing handwritten Bengali characters. Since Bengali consists of differ- Journal homepage: http://ijai.iaescore.com