Comparative analysis of machine learning approaches in Kazakh banknote classification

Aug 5, 2026

Upload by Miles

Research presents a smartphone-camera based system for classifying Kazakh banknotes, targeting users including people with visual impairments. The study builds lightweight and accurate classifiers for 500 KZT–20,000 KZT images and compares traditional machine learning with a hybrid pipeline using feature extraction from a pre-trained ResNet-18. A dataset of 4,200 images with realistic variations supports evaluation. Results show 94.00% accuracy for the traditional model, 99.11% for the hybrid model, and model stacking reaching 99.55%.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Introduction

Motivation and inclusivity

Traditional vs hybrid ML approaches

Dataset and denomination scope

Methodology and evaluation

Classifier design and feature extraction

Optimization and model stacking

Metrics and comparative results

Transferability and discussion

Extension to other image tasks

Comparative analysis of machine learning approaches in Kazakh banknote classification Ualikhan Sadyk, Makhambet Yerzhan, Rashid Baimukashev, Cemil Turan Department of Сomputer Science, Faculty of Engineering and Natural Science, SDU University, Kaskelen, Kazakhstan Article history: Received Dec 9, 2023 Revised Mar 14, 2024 Accepted Mar 28, 2024 Keywords: Banknote classification Hybrid approach Kazakh banknotes Machine learning Model stacking Traditional approach Visual impairments Corresponding Author: Nowadays, smartphones seamlessly blend into every aspect of our lives, including as handheld assistants for individuals with disabilities. Therefore, this research addresses the need for a robust system that can classify Kazakh banknotes. By capitalizing on the availability of smartphones and the ability to integrate detectors with classifiers this study introduces classifiers of Kazakh banknote images specifically designed for banknotes ranging from 500 KZT to 20,000 KZT. It compares traditional and hybrid machine learning (ML) approaches, utilizing a dataset of diverse banknote images, aiming for both lightweight and high accuracy. Competitive performance is demonstrated by the traditional approach, enhanced by thoughtful feature engineering. The hybrid approach, utilizing features from a pre-trained ResNet-18 model, showcases remarkable accuracy and robustness. Evaluation metrics reveal significant achievements, with the traditional approach attaining 94.00% accuracy and the hybrid approach excelling at 99.11% . Model stacking, combining classifiers from both approaches, outperforms individual classifiers, achieving 95.00% and 99.55% accuracy for the traditional and hybrid ML approaches, respectively. Our methodology’s comparable outcome in classifying Thai banknotes and coffee beans roasting levels demonstrates their versatility in image classification tasks that rely on color differentiation, showcasing the potential beyond banknote recognition. This is an open access article under the CC BY-SA license. Ualikhan Sadyk Department of Сomputer Science, Faculty of Engineering and Natural Science SDU University Abylaikhan St. 1/1, Kaskelen, Kazakhstan [Email: ualikhan.sadyk@sdu.edu.kz](Email: ualikhan.sadyk@sdu.edu.kz) Article Info ABSTRACT 1. INTRODUCTION In an era characterized by technological advancements and a heightened emphasis on inclusivity, the creation of efficient and accurate systems to assist individuals with visual impairments is deemed crucial. With the prevalence of smartphones equipped with high-quality cameras in today’s society being recognized, a mission is undertaken to address the specific need of classifying Kazakh banknotes [1] . Leveraging the ubiquity of smartphones, which serve as indispensable tools in daily lives, the aim of this study is to pioneer a сlassifier of Kazakh banknote images. The primary objective is twofold: the creation of a lightweight and highly accurate Kazakh banknote classifier, and the comparison of the efficacy of traditional and hybrid machine learning (ML) approaches in achieving this goal. Traditional ML approaches, rooted in manual feature extraction and well-established ML techniques, offer interpretability and control [2] . In contrast, hybrid ML approaches, integrating automated feature extraction from pre-trained deep learning models, bring forth the potential to capture intricate patterns and nuances within banknote images [3], [4] . The employed dataset, comprising 4,200 high-quality images of Kazakhstani banknotes under diverse conditions–varying lighting environments, cluttered backgrounds, and folded banknotes–mimics the challenges encountered in everyday life [5] . By narrowing the scope to the denominations commonly utilized, ranging from 500 KZT to 20000 KZT, the aim is to create a classifier that aligns with practical, realworld scenarios. The potential impact of this research extends beyond technical innovation. By empow

FAQs

What problem does the research address in Kazakh banknote classification?

The work targets the need for an accurate, robust Kazakh banknote classification system that can support individuals with visual impairments using smartphones with high-quality cameras.

How do the traditional and hybrid machine learning approaches differ?

Traditional approaches rely on manual feature extraction and established ML methods, while the hybrid approach uses automated feature extraction from a pre-trained ResNet-18 model to capture patterns in banknote images.

What accuracy results are reported for individual models and model stacking?

The traditional approach achieves 94.00% accuracy and the hybrid approach reaches 99.11%. Model stacking combining both approaches performs best, reaching 99.55% accuracy for the hybrid combined setup.

Aug 5, 2026

Upload by Miles