Introducing Image Classification through No-Code Teachable Machine Platform: An Intuitive Approach for Novice Learners

Introducing Image Classification through No-Code Teachable Machine Platform: An Intuitive Approach for Novice Learners | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Method Article Introducing Image Classification through No-Code Teachable Machine Platform: An Intuitive Approach for Novice Learners Ritu Raj Lamsal, Skand Lamsal This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7625905/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The technical complexity associated with traditional machine learning (ML) tools often impedes early-stage learners and non-programmers from exploring artificial intelligence (AI). This paper presents a pedagogically intuitive approach using Google’s Teachable Machine a no-code platform that allows users to create ML models via a graphical interface. We conducted an educational experiment where novice learners built an image classifier to distinguish between animal classes. Our experiment involved the classification of four animal class Cat, Dog, Monkey, and Tiger using a minimal dataset of just 3–5 images per class to ensure simplicity for beginners. Using transfer learning with MobileNet architecture and basic training settings (50 epochs, 0.001 learning rate), the model achieved high classification accuracy 100% for visually distinct classes and up to 97% for more similar ones. The platform’s ability to export trained models to TensorFlow, TensorFlow Lite, expands its potential for real-world deployment. This study concludes that GUI-based ML tools can democratize AI education and discuss their integration with IoT applications, making it accessible, interactive, and impactful for beginners. Artificial Intelligence and Machine Learning Teachable Machine Machine Learning for Beginners Image Classification IoT and ML No-Code ML MobileNet TensorFlow Figures Figure 1 Figure 2 Figure 3 1. Introduction The global proliferation of Artificial Intelligence (AI) and Machine Learning (ML) is transforming nearly every sector—from healthcare and finance to education and agriculture. These technologies are not only enabling automation and efficiency but also allowing for entirely new ways of solving traditional problems through data-driven insights. The rapid growth in Internet of Things (IoT) and integration with AI has a huge impact and paradigm shift in many sectors [ 1 ]. However, while AI and ML have become dominant forces in modern innovation, understanding and implementing these technologies remain challenging, especially for novice learners with limited technical backgrounds. Traditional ML education involves complex concepts such as algorithm selection, feature extraction, model tuning, and programming, which can be overwhelming for beginners [ 2 ]. Tools like TensorFlow, PyTorch, or Scikit-learn are powerful but demand prior knowledge of Python and foundational understanding of linear algebra, calculus, and statistics [ 3 ]. This learning curve has often created a digital divide limiting AI exposure to only those with strong computational skills or access to formal STEM education. To address this barrier, simplified, visual, and no-code platforms have emerged as effective educational tools. Google’s Teachable Machine [ 4 ] is one such initiative that empowers learners to create custom ML models for image, sound, or pose recognition without writing a single line of code. Designed with accessibility and privacy in mind, Teachable Machine supports both file uploads and live input through webcams and microphones, and allows the resulting models to be deployed entirely on-device. This flexibility makes it a suitable tool for classrooms, individual learners, and even rural or low-resource settings. This paper presents an educational demonstration using Teachable Machine to train an image classification model that can distinguish between four animal classes cat, dog, monkey, and tiger using just 3–5 images per class. The minimal dataset is intentional, showing that even with limited data, one can build a meaningful model using transfer learning. By visualizing the process of model training, prediction, and accuracy evaluation, learners can gain practical insights into ML fundamentals without being overwhelmed by coding or mathematical abstraction. The importance of such accessible tools becomes even more evident when considered in the context of Internet of Things (IoT) applications. IoT devices generate real-time data that often requires intelligent analysis for actionable insights. Integrating ML into IoT systems allows for predictive analytics, anomaly detection, and autonomous decision-making—paving the way for smarter cities, precision agriculture, remote health monitoring, and more [ 4 ][ 5 ][ 6 ][ 7 ]. By introducing ML through platforms like Teachable Machine, even non-engineers such as farmers, school students, and local entrepreneurs can begin to understand how smart technologies work and how they can be applied in their own contexts. Moreover, the simplicity of the interface makes it possible to incorporate such tools in community training, technical education programs, and innovation hubs, particularly in low- and middle-income countries. The integration of the Internet of Things (IoT) and machine learning (ML) in education has the potential to transform traditional methods of classroom instruction into digitalized instruction methods [ 8 ]. In summary, this study not only provides a gateway for students to learn about ML through image classification but also draws connections to how such models can be deployed in real-world settings, including IoT-based systems [ 9 ][ 10 ]. Our goal is to foster curiosity, lower entry barriers, and ultimately promote responsible and inclusive AI education. The remaining part of the paper is organized as follows. Related work is highlighted in section 2 . Section 3 discusses the methodology. Results and discussion are presented in section 4 , followed by conclusion and future work in section 5 and 6 respectively. 2. Related Works Image classification is a fundamental task in computer vision, with applications across industries such as healthcare, autonomous driving, and entertainment. Over the years, several machine learning algorithms have been developed to solve this challenge, each offering distinct advantages and limitations. Traditional machine learning methods for image classification typically involve complex concepts such as feature extraction, algorithm selection, and model tuning, which can be overwhelming for beginners. Tools like TensorFlow , PyTorch , and Scikit-learn are powerful but demand prior knowledge of Python and foundational concepts in linear algebra, calculus, and statistics [ 11 ]. This steep learning curve often limits AI education to those with strong computational skills or access to formal STEM education, creating a digital divide that hinders broad access to machine learning and artificial intelligence technologies. A major leap in image classification came with AlexNet by Krizhevsky et al. [ 12 ], which utilized an eight-layer deep convolutional neural network (CNN) and achieved a top-5 error rate of 15.3% on ImageNet. This success demonstrated the potential of deep learning and led to even deeper models like VGGNet , which standardized architecture using 16 and 19 layers and reduced the top-5 error rate to 7.3%, albeit at the cost of significantly more parameters (~ 138 million) [ 13 ]. The issue of vanishing gradients in very deep networks was addressed by ResNet , which introduced residual connections that facilitated the training of models with over 100 layers. ResNet-152 achieved a top-5 error rate of 3.57%, setting a new benchmark for performance while demonstrating the effectiveness of residual learning [ 14 ]. Following this, DenseNet introduced dense connectivity patterns where each layer received inputs from all preceding layers, promoting feature reuse and improved parameter efficiency. DenseNet-201 , for example, achieved competitive performance with only ~ 20 million parameters [ 15 ]. For deployment in resource-constrained environments, SqueezeNet provided an efficient alternative, reaching AlexNet-level accuracy with just 1.2 million parameters. This was achieved using “fire modules,” which utilized 1×1 convolutions to drastically reduce parameter count without significant accuracy loss [ 16 ]. The introduction of the Vision Transformer (ViT) marked a shift away from CNNs by adapting transformer models originally developed for natural language processing to vision tasks. ViT partitions input images into patches and processes them as token sequences using self-attention mechanisms, achieving state-of-the-art performance on large datasets like ImageNet [ 19 ]. However, ViT requires substantial computational resources and large-scale pretraining, making it less practical in constrained settings. To address these limitations, the Swin Transformer introduced a hierarchical design with shifted window attention, which enables both local and global feature modeling with greater efficiency. It has proven effective not only in classification but also in detection and segmentation tasks [ 20 ]. A complementary approach focused on efficiency is the EfficientNet family, which employs compound scaling to jointly optimize network depth, width, and resolution. EfficientNet-B7 reached top-tier accuracy on ImageNet with far fewer parameters and FLOPs than many previous models [ 17 ]. Its successor, EfficientNetV2 , further improved training speed and generalization, especially on smaller datasets, through fused layers and progressive learning strategies [ 18 ]. Finally, ConvNeXt emerged as a modernized convolutional architecture that borrows design innovations from transformers (such as GELU activation, layer normalization, and inverted bottlenecks) while maintaining the simplicity and efficiency of CNNs. It achieves competitive performance with ViTs, reaffirming the relevance of convolutions in the era of attention-based models [ 21 ]. Table 1 Performance summary of discussed model Model Year Top-1 Accuracy Parameters Key Features AlexNet 2012 57.2% 60M First deep CNN for ImageNet; ReLU activation; dropout for regularization VGG-16 2014 71.5% 138M Uniform layer structure; large parameter count ResNet-152 2015 78.6% 60M Residual (skip) connections; enables deeper networks DenseNet-201 2017 77.3% 20M Dense connectivity; improves feature reuse and efficiency SqueezeNet 2016 57.5% 1.2M Compact model; good for edge and embedded devices MobileNetV2 2018 72.0% 3.4M Depthwise separable convolutions; optimized for mobile and embedded use EfficientNet-B0 2019 77.1% 5.3M Compound scaling of depth, width, and resolution for better efficiency Vision Transformer (ViT-B/16) 2020 85.2% 86M Transformer-based; processes image patches like tokens; needs large datasets Swin Transformer 2021 84.5% 87M Hierarchical transformer with shifted windows; excels in vision tasks ConvNeXt-T 2022 82.1% 28M CNN redesigned to match transformer performance; inspired by ResNet EfficientNetV2-S 2021 84.6% 22M Improved training speed and parameter efficiency over original EfficientNet These advancements have pushed the boundaries of what is possible in image classification, but the associated computational complexity poses a barrier for educational outreach, particularly for novice learners. Platforms like Teachable Machine make machine learning more accessible by providing a simple, intuitive interface for creating models with minimal data and computational resources, helping to bridge the gap between curiosity and competency in AI education. 3. Methodology This study adopted an experimental and educational methodology centered on Google’s Teachable Machine. The primary objective was to create a simplified yet effective image classification model that beginners can replicate. The methodology is divided into four stages: data preparation, model training, evaluation, and export. Figure 1 shows a no-code model for image classification. 3.1 Data Preparation To reduce complexity and keep the process learner-friendly, we chose four image categories: cat, dog, monkey, and tiger. For each class, only 3 to 5 images were selected as seen in Fig. 2 . The images were either uploaded from local storage or captured live using the webcam interface within Teachable Machine. This deliberate limitation was intended to show that even with a small dataset, a model can be successfully trained using transfer learning. 3.2 Model Training Teachable Machine uses the MobileNet architecture, which is optimized for mobile and edge devices. We trained the model for 50 epochs with a learning rate of 0.001. The platform manages preprocessing and model configuration automatically, allowing users to focus on understanding the classification process without delving into technicalities like data augmentation or neural network architecture. 3.3 Evaluation and Export After training, the model’s performance was tested using webcam input and static test images. The platform allows the model to be exported in both TensorFlow and TensorFlow Lite format. These models can then be used in custom applications or integrated with IoT systems for real-time inference. The platform also supports on-device usage, meaning no webcam or microphone data is transmitted to external servers ensuring user privacy and aligning with ethical AI design principles. 4. Results and Discussion Despite the minimal dataset size, the model showed surprisingly strong performance in distinguishing between the selected classes. During testing, the system achieved 100% classification accuracy in easily distinguishable pairs such as cats vs. dogs and dogs vs. tigers. More nuanced comparisons, such as cats vs. monkeys or monkeys vs. tigers, still yielded high accuracy rates of 97% and 98%, respectively as depicted in Fig. 3 . These results underscore the effectiveness of transfer learning, even with limited data, and confirm that Teachable Machine is robust enough to handle basic classification tasks with minimal user input. One of the most valuable aspects of this experiment was observing how quickly novice users could understand and engage with ML principles. Learners could visualize how different inputs (images) were being associated with specific labels and how the model refined its predictions over time during training. This hands-on, visual learning experience demystifies the concept of machine learning, making it more accessible and less intimidating for those without a programming background. Another important discussion point is the tool’s versatility. While this study focused on image classification, Teachable Machine also supports sound (audio classification) and pose (body position recognition), expanding its educational utility across different types of sensory data. Beginners can create models that recognize musical notes, spoken words, or physical gestures simply by recording examples through a microphone or webcam. This multi-modal capability encourages creativity and interdisciplinary applications, further enhancing its value in educational settings. The trained image classification model performed remarkably well despite the limited training data. For visually distinct classes, such as dogs and tigers, accuracy reached 100%. For more nuanced distinctions, such as between monkeys and cats, accuracy still remained above 97%, highlighting the robustness of MobileNet’s pre-trained weights and the effectiveness of transfer learning. These results are significant from an educational perspective. Novice learners, including school students and early-stage professionals, can clearly see the impact of labeled data on model accuracy. They are introduced to the concept of overfitting, training epochs, and prediction confidence, all within a simple, visual interface. By encouraging active participation e.g., by allowing users to record their own images Teachable Machine fosters experiential learning. Moreover, the application potential goes beyond image classification. Learners can use the same platform to build models that recognize sound (e.g., voice commands, musical instruments) or body poses (e.g., hand signals, yoga positions). These applications have clear relevance to IoT systems where such inputs might control devices or trigger alerts. The model’s ability to be exported and deployed across environments web applications, mobile apps, or embedded devices mirrors the practical deployment of ML in IoT-based projects. These integrations underscore the real-world applicability of concepts learned through Teachable Machine, connecting ML education directly to societal use cases. 5. Conclusion This study demonstrates that Teachable Machine serves as an effective tool for introducing machine learning to a wide audience. By enabling image classification with only a few examples per class, the platform shows how transfer learning can simplify model training for beginners. Through hands-on experimentation, learners can understand the lifecycle of an ML model from data gathering to deployment without needing to write code. The flexible export options and privacy-preserving design of Teachable Machine further enhance its suitability for education. It can serve as a gateway to more advanced ML tools and, eventually, to integrated systems involving IoT and real-time automation. For institutions and educators looking to integrate AI into their curriculum or outreach, Teachable Machine offers a powerful starting point. 6. Future Work While this project focused on a basic image classification task with four animal classes, future work can explore additional data modalities and applications. For instance, sound classification could help learners understand how audio features contribute to recognition models. Pose classification could be used in health monitoring or gesture-based device control. Another promising area is the integration of Teachable Machine with physical computing platforms such as Arduino or Raspberry Pi, thereby linking ML models with sensor-driven IoT systems. This would enable learners to simulate real-world smart applications such as a model that uses image data to automate plant disease detection [ 22 ], face recognition, home security and automation and many more. Finally, incorporating Teachable Machine into formal education through structured lesson plans and projects aligned with local challenges would promote contextual learning and innovation among youth and grassroots communities. Declarations Contribution : Ritu R Lamsal involved in Conceptualization, manuscript drafting, literature review and S Lamsal contributed in data /sample collection, model training and review writing. Conflict of Interest : Authors declares no conflict of interest. Funding : No external funding Acknowledgement: The authors would like to thank all the students who encourage us to work on this article. We also extend our sincere Acknowledgement to Google and teachable machine platform. References Lamsal RR, Bhattarai M, Lamsal S (2025) AI and IoT Integration in Engineering: Use Cases, Challenges, and Opportunities in Nepal, Jan. [Online]. Available: https://doi.org/10.13140/RG.2.2.30872.51208 Thornton C, Hutter F, Hoos HH, Leyton-Brown K Auto-WEKA: Combined selection and hyperparameter optimization of classification algorithms, in *Proc. 19th ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining*, 2013, pp. 847–855. [Online]. Available: https://doi.org/10.1145/2487575.2487629 Raschka S, Patterson J, Nolet C Machine learning in Python: Main developments and technology trends in data science, machine learning, and artificial intelligence, *Information*, vol. 11, no. 4, p. 193, 2020. [Online]. Available: https://doi.org/10.3390/info11040193 Google (2024) Teachable Machine: Train a computer to recognize your own images, sounds, & poses, [Online]. Available: https://teachablemachine.withgoogle.com Lamsal RR, Karthikeyan P, Otero P, Ariza A (1900) Design and implementation of Internet of Things (IoT) platform targeted for smallholder farmers: From Nepal perspective, *Agriculture*, vol. 13, no. 10, p. 2023. [Online]. Available: https://doi.org/10.3390/agriculture13101900 Lamsal RR, Acharya UK, Karthikeyan P, Otero P, Ariza Quintana A (2024) Implementing Internet of Things for real-time monitoring and regulation in off-season grafting and post-harvest storage in citrus cultivation: A case study from the hilly regions of Nepal. *Manuscript submitted for publication* R. R. Lamsal *et al*., Monitoring and regulating climatic condition of polyhouse for successful off-season grafting of citrus fruits using Internet of Things platform, *Int. J. Agric. Environ. Biotechnol.*, vol. 15, Special Issue, pp. 417–422, (2022) [Online]. Available: https://aiasa.org.in/wp-content/uploads/2022/09/IJAEB-Sp-Issue-AAFS-2022-1-1.pdf M. A. Almaiah *et al*., Machine learning adoption in educational institutions: Role of Internet of Things and digital educational platforms, *Sustainability*, vol. 15, no. 5, p. 4000, (2023) A. G. Howard *et al*., MobileNets: Efficient convolutional neural networks for mobile vision applications, *arXiv preprint*, arXiv:1704.04861, 2017. [Online]. Available: https://arxiv.org/abs/1704.04861 Chen D, Chen Y, Li Y, Mao F, He Y, Xue H (2021), June Self-supervised learning for few-shot image classification. In ICASSP 2021–2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 1745–1749). IEEE Goodfellow I, Bengio Y, Courville A, *Deep Learning*, Cambridge MA (2016) MIT Press, [Online]. Available: http://alvarestech.com/temp/deep/Deep%20Learning%20by%20Ian%20Goodfellow,%20Yoshua%20Bengio,%20Aaron%20Courville%20 (z-lib.org).pdf [Accessed: May 10, 2025] Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural networks, in *Advances in Neural Inf. Process. Syst.*, vol. 25, pp. 1097–1105 Simonyan K, Zisserman A Very deep convolutional networks for large-scale image recognition, *arXiv preprint*, arXiv:1409.1556, 2014. [Online]. Available: https://arxiv.org/abs/1409.1556 He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition, in *Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR)*, pp. 770–778 Huang G, Liu Z, van der Maaten L, Weinberger KQ (2017) Densely connected convolutional networks, in *Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR)*, pp. 4700–4708 Iandola FN, Moskewicz MW, Ashraf K, Golz M, Keutzer K SqueezeNet: AlexNet-level accuracy with 50× fewer parameters and < 0.5MB model size, *arXiv preprint*, arXiv:1602.07360, 2016. [Online]. Available: https://arxiv.org/abs/1602.07360 Tan M, Le QV EfficientNet: Rethinking model scaling for convolutional neural networks, in *Proc. Int. Conf. Mach. Learn. (ICML)*, 2019, pp. 6105–6114. [Online]. Available: https://arxiv.org/abs/1905.11946 Tan M, Le QV EfficientNetV2: Smaller models and faster training, in *Proc. Int. Conf. Mach. Learn. (ICML)*, 2021. [Online]. Available: https://arxiv.org/abs/2104.0029 Dosovitskiy *et A al*., An image is worth 16×16 words: Transformers for image recognition at scale, in *Proc. Int. Conf. Learn. Represent. (ICLR)*, 2020. [Online]. Available: https://arxiv.org/abs/2010.11929 Liu *et al* Z et al (2021) Swin Transformer: Hierarchical vision transformer using shifted windows, in *Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV)*, pp. 10012–10022. [Online]. Available: https://arxiv.org/abs/2103.14030 Liu *et al*. Z et al A ConvNet for the 2020s, in *Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR)*, 2022, pp. 11976–11986. [Online]. Available: https://arxiv.org/abs/2201.03545 Lamsal RR, Acharya P, Pokhrel R et al Integrating Machine Learning and RAG-Based Chatbot for Mandarin Orange Disease Detection in Hilly Region of Nepal, 27 February 2025, PREPRINT (Version 1) available at Research Square [ https://doi.org/10.21203/rs.3.rs-6105402/v1] Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7625905","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Method Article","associatedPublications":[],"authors":[{"id":515675339,"identity":"2a3a99ca-20b8-498e-8608-be4c8bfe380d","order_by":0,"name":"Ritu Raj Lamsal","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3klEQVRIiWNgGAWjYBACPjBZACIOHwASEjIEtbCBSQMQcSwBpIWHFC08EJKwFvbTyR9+GDDk8zee+fzqRo0FDwP74aMb8Grhyd0m2WPAYDnjwNlt1jnHgA7jSUu7gd9hudtATjJgAGoxzmEDapHgMcOvhf/t5o9/gFrkD5x5ZpzzjxgtErkbpEG2GBw4w/w4t40oLW+3ScsYSBgYHjhmxpzbJ8HDRsgv/Py5mz++qbAxkLtx+PHnnG91cvzsh4/h1QIFEkB0gE0CbC8RymH2NTB/IF71KBgFo2AUjCQAAIVOQznoNQAOAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-3751-2785","institution":"Madan Bhandari University of Science and Technology","correspondingAuthor":true,"prefix":"","firstName":"Ritu","middleName":"Raj","lastName":"Lamsal","suffix":""},{"id":515675340,"identity":"eebd9c9d-f22a-498d-b618-56e9eefe0530","order_by":1,"name":"Skand Lamsal","email":"","orcid":"","institution":"St Xaviers College , Maitighar, Nepal","correspondingAuthor":false,"prefix":"","firstName":"Skand","middleName":"","lastName":"Lamsal","suffix":""}],"badges":[],"createdAt":"2025-09-16 04:59:54","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7625905/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7625905/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91526640,"identity":"e674aef1-40b0-47b1-b22e-f449db853f11","added_by":"auto","created_at":"2025-09-17 11:10:38","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":33589,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eNo-Code Model for Image Classification\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7625905/v1/96b198120659489006430e96.png"},{"id":91524794,"identity":"55a7c896-2f49-424a-8c2d-d7d9e7bb9971","added_by":"auto","created_at":"2025-09-17 11:02:38","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":264557,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSample Images for training and test\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7625905/v1/0939bb10c6f347a1fee8573f.png"},{"id":91524789,"identity":"df189a56-b132-4c17-84c1-ff0f6891dd6c","added_by":"auto","created_at":"2025-09-17 11:02:38","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":355199,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(a) Classification and accuracy of Tiger (b) Classification and accuracy of Cat\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(c) Classification and accuracy of Dog (d) Classification and accuracy of Monkey\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7625905/v1/c73a1df491c03f5d1b48d2fc.jpeg"},{"id":91526642,"identity":"92b4d3c7-8f96-49a1-8008-47470e97c199","added_by":"auto","created_at":"2025-09-17 11:10:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1335126,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7625905/v1/525481e8-3cfb-4816-9f72-a0dd1b5b85fc.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eIntroducing Image Classification through No-Code Teachable Machine Platform: An Intuitive Approach for Novice Learners\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe global proliferation of Artificial Intelligence (AI) and Machine Learning (ML) is transforming nearly every sector\u0026mdash;from healthcare and finance to education and agriculture. These technologies are not only enabling automation and efficiency but also allowing for entirely new ways of solving traditional problems through data-driven insights. The rapid growth in Internet of Things (IoT) and integration with AI has a huge impact and paradigm shift in many sectors [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. However, while AI and ML have become dominant forces in modern innovation, understanding and implementing these technologies remain challenging, especially for novice learners with limited technical backgrounds. Traditional ML education involves complex concepts such as algorithm selection, feature extraction, model tuning, and programming, which can be overwhelming for beginners [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Tools like TensorFlow, PyTorch, or Scikit-learn are powerful but demand prior knowledge of Python and foundational understanding of linear algebra, calculus, and statistics [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. This learning curve has often created a digital divide limiting AI exposure to only those with strong computational skills or access to formal STEM education.\u003c/p\u003e\u003cp\u003eTo address this barrier, simplified, visual, and no-code platforms have emerged as effective educational tools. Google\u0026rsquo;s Teachable Machine [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] is one such initiative that empowers learners to create custom ML models for image, sound, or pose recognition without writing a single line of code. Designed with accessibility and privacy in mind, Teachable Machine supports both file uploads and live input through webcams and microphones, and allows the resulting models to be deployed entirely on-device. This flexibility makes it a suitable tool for classrooms, individual learners, and even rural or low-resource settings. This paper presents an educational demonstration using Teachable Machine to train an image classification model that can distinguish between four animal classes cat, dog, monkey, and tiger using just 3\u0026ndash;5 images per class. The minimal dataset is intentional, showing that even with limited data, one can build a meaningful model using transfer learning. By visualizing the process of model training, prediction, and accuracy evaluation, learners can gain practical insights into ML fundamentals without being overwhelmed by coding or mathematical abstraction.\u003c/p\u003e\u003cp\u003eThe importance of such accessible tools becomes even more evident when considered in the context of Internet of Things (IoT) applications. IoT devices generate real-time data that often requires intelligent analysis for actionable insights. Integrating ML into IoT systems allows for predictive analytics, anomaly detection, and autonomous decision-making\u0026mdash;paving the way for smarter cities, precision agriculture, remote health monitoring, and more [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e][\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e][\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e][\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. By introducing ML through platforms like Teachable Machine, even non-engineers such as farmers, school students, and local entrepreneurs can begin to understand how smart technologies work and how they can be applied in their own contexts. Moreover, the simplicity of the interface makes it possible to incorporate such tools in community training, technical education programs, and innovation hubs, particularly in low- and middle-income countries. The integration of the Internet of Things (IoT) and machine learning (ML) in education has the potential to transform traditional methods of classroom instruction into digitalized instruction methods [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn summary, this study not only provides a gateway for students to learn about ML through image classification but also draws connections to how such models can be deployed in real-world settings, including IoT-based systems [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e][\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Our goal is to foster curiosity, lower entry barriers, and ultimately promote responsible and inclusive AI education. The remaining part of the paper is organized as follows. Related work is highlighted in section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Section \u003cspan refid=\"Sec3\" class=\"InternalRef\"\u003e3\u003c/span\u003e discusses the methodology. Results and discussion are presented in section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e4\u003c/span\u003e, followed by conclusion and future work in section \u003cspan refid=\"Sec8\" class=\"InternalRef\"\u003e5\u003c/span\u003e and \u003cspan refid=\"Sec9\" class=\"InternalRef\"\u003e6\u003c/span\u003e respectively.\u003c/p\u003e"},{"header":"2. Related Works","content":"\u003cp\u003eImage classification is a fundamental task in computer vision, with applications across industries such as healthcare, autonomous driving, and entertainment. Over the years, several machine learning algorithms have been developed to solve this challenge, each offering distinct advantages and limitations.\u003c/p\u003e\u003cp\u003eTraditional machine learning methods for image classification typically involve complex concepts such as feature extraction, algorithm selection, and model tuning, which can be overwhelming for beginners. Tools like \u003cb\u003eTensorFlow\u003c/b\u003e, \u003cb\u003ePyTorch\u003c/b\u003e, and \u003cb\u003eScikit-learn\u003c/b\u003e are powerful but demand prior knowledge of Python and foundational concepts in linear algebra, calculus, and statistics [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. This steep learning curve often limits AI education to those with strong computational skills or access to formal STEM education, creating a digital divide that hinders broad access to machine learning and artificial intelligence technologies.\u003c/p\u003e\u003cp\u003eA major leap in image classification came with \u003cb\u003eAlexNet\u003c/b\u003e by Krizhevsky \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], which utilized an eight-layer deep convolutional neural network (CNN) and achieved a top-5 error rate of 15.3% on ImageNet. This success demonstrated the potential of deep learning and led to even deeper models like \u003cb\u003eVGGNet\u003c/b\u003e, which standardized architecture using 16 and 19 layers and reduced the top-5 error rate to 7.3%, albeit at the cost of significantly more parameters (~\u0026thinsp;138\u0026nbsp;million) [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe issue of vanishing gradients in very deep networks was addressed by \u003cb\u003eResNet\u003c/b\u003e, which introduced residual connections that facilitated the training of models with over 100 layers. \u003cb\u003eResNet-152\u003c/b\u003e achieved a top-5 error rate of 3.57%, setting a new benchmark for performance while demonstrating the effectiveness of residual learning [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Following this, \u003cb\u003eDenseNet\u003c/b\u003e introduced dense connectivity patterns where each layer received inputs from all preceding layers, promoting feature reuse and improved parameter efficiency. \u003cb\u003eDenseNet-201\u003c/b\u003e, for example, achieved competitive performance with only\u0026thinsp;~\u0026thinsp;20\u0026nbsp;million parameters [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFor deployment in resource-constrained environments, \u003cb\u003eSqueezeNet\u003c/b\u003e provided an efficient alternative, reaching AlexNet-level accuracy with just 1.2\u0026nbsp;million parameters. This was achieved using \u0026ldquo;fire modules,\u0026rdquo; which utilized 1\u0026times;1 convolutions to drastically reduce parameter count without significant accuracy loss [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eThe introduction of the \u003cb\u003eVision Transformer (ViT)\u003c/b\u003e marked a shift away from CNNs by adapting transformer models originally developed for natural language processing to vision tasks. ViT partitions input images into patches and processes them as token sequences using self-attention mechanisms, achieving state-of-the-art performance on large datasets like ImageNet [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. However, ViT requires substantial computational resources and large-scale pretraining, making it less practical in constrained settings.\u003c/p\u003e\u003cp\u003eTo address these limitations, the \u003cb\u003eSwin Transformer\u003c/b\u003e introduced a hierarchical design with shifted window attention, which enables both local and global feature modeling with greater efficiency. It has proven effective not only in classification but also in detection and segmentation tasks [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eA complementary approach focused on efficiency is the \u003cb\u003eEfficientNet\u003c/b\u003e family, which employs compound scaling to jointly optimize network depth, width, and resolution. \u003cb\u003eEfficientNet-B7\u003c/b\u003e reached top-tier accuracy on ImageNet with far fewer parameters and FLOPs than many previous models [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Its successor, \u003cb\u003eEfficientNetV2\u003c/b\u003e, further improved training speed and generalization, especially on smaller datasets, through fused layers and progressive learning strategies [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eFinally, \u003cb\u003eConvNeXt\u003c/b\u003e emerged as a modernized convolutional architecture that borrows design innovations from transformers (such as GELU activation, layer normalization, and inverted bottlenecks) while maintaining the simplicity and efficiency of CNNs. It achieves competitive performance with ViTs, reaffirming the relevance of convolutions in the era of attention-based models [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance summary of discussed model\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYear\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTop-1 Accuracy\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eParameters\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eKey Features\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAlexNet\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2012\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e57.2%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e60M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eFirst deep CNN for ImageNet; ReLU activation; dropout for regularization\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVGG-16\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2014\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e71.5%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e138M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eUniform layer structure; large parameter count\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eResNet-152\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2015\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e78.6%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e60M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eResidual (skip) connections; enables deeper networks\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eDenseNet-201\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2017\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e77.3%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e20M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDense connectivity; improves feature reuse and efficiency\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eSqueezeNet\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2016\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e57.5%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.2M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCompact model; good for edge and embedded devices\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eMobileNetV2\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2018\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e72.0%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3.4M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eDepthwise separable convolutions; optimized for mobile and embedded use\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eEfficientNet-B0\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2019\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e77.1%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e5.3M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCompound scaling of depth, width, and resolution for better efficiency\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eVision Transformer (ViT-B/16)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2020\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e85.2%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e86M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eTransformer-based; processes image patches like tokens; needs large datasets\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eSwin Transformer\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2021\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e84.5%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e87M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHierarchical transformer with shifted windows; excels in vision tasks\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eConvNeXt-T\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2022\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e82.1%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e28M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eCNN redesigned to match transformer performance; inspired by ResNet\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eEfficientNetV2-S\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e2021\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e84.6%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e22M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eImproved training speed and parameter efficiency over original EfficientNet\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThese advancements have pushed the boundaries of what is possible in image classification, but the associated computational complexity poses a barrier for educational outreach, particularly for novice learners. Platforms like Teachable Machine make machine learning more accessible by providing a simple, intuitive interface for creating models with minimal data and computational resources, helping to bridge the gap between curiosity and competency in AI education.\u003c/p\u003e"},{"header":"3. Methodology","content":"\u003cp\u003eThis study adopted an experimental and educational methodology centered on Google\u0026rsquo;s Teachable Machine. The primary objective was to create a simplified yet effective image classification model that beginners can replicate. The methodology is divided into four stages: data preparation, model training, evaluation, and export. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows a no-code model for image classification.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Data Preparation\u003c/h2\u003e\u003cp\u003eTo reduce complexity and keep the process learner-friendly, we chose four image categories: cat, dog, monkey, and tiger. For each class, only 3 to 5 images were selected as seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The images were either uploaded from local storage or captured live using the webcam interface within Teachable Machine. This deliberate limitation was intended to show that even with a small dataset, a model can be successfully trained using transfer learning.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Model Training\u003c/h2\u003e\u003cp\u003eTeachable Machine uses the MobileNet architecture, which is optimized for mobile and edge devices. We trained the model for 50 epochs with a learning rate of 0.001. The platform manages preprocessing and model configuration automatically, allowing users to focus on understanding the classification process without delving into technicalities like data augmentation or neural network architecture.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Evaluation and Export\u003c/h2\u003e\u003cp\u003eAfter training, the model\u0026rsquo;s performance was tested using webcam input and static test images. The platform allows the model to be exported in both TensorFlow and TensorFlow Lite format. These models can then be used in custom applications or integrated with IoT systems for real-time inference. The platform also supports on-device usage, meaning no webcam or microphone data is transmitted to external servers ensuring user privacy and aligning with ethical AI design principles.\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Results and Discussion","content":"\u003cp\u003eDespite the minimal dataset size, the model showed surprisingly strong performance in distinguishing between the selected classes. During testing, the system achieved 100% classification accuracy in easily distinguishable pairs such as cats vs. dogs and dogs vs. tigers. More nuanced comparisons, such as cats vs. monkeys or monkeys vs. tigers, still yielded high accuracy rates of 97% and 98%, respectively as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003e. These results underscore the effectiveness of transfer learning, even with limited data, and confirm that Teachable Machine is robust enough to handle basic classification tasks with minimal user input.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eOne of the most valuable aspects of this experiment was observing how quickly novice users could understand and engage with ML principles. Learners could visualize how different inputs (images) were being associated with specific labels and how the model refined its predictions over time during training. This hands-on, visual learning experience demystifies the concept of machine learning, making it more accessible and less intimidating for those without a programming background. Another important discussion point is the tool\u0026rsquo;s versatility. While this study focused on image classification, Teachable Machine also supports sound (audio classification) and pose (body position recognition), expanding its educational utility across different types of sensory data. Beginners can create models that recognize musical notes, spoken words, or physical gestures simply by recording examples through a microphone or webcam. This multi-modal capability encourages creativity and interdisciplinary applications, further enhancing its value in educational settings.\u003c/p\u003e\u003cp\u003eThe trained image classification model performed remarkably well despite the limited training data. For visually distinct classes, such as dogs and tigers, accuracy reached 100%. For more nuanced distinctions, such as between monkeys and cats, accuracy still remained above 97%, highlighting the robustness of MobileNet\u0026rsquo;s pre-trained weights and the effectiveness of transfer learning. These results are significant from an educational perspective. Novice learners, including school students and early-stage professionals, can clearly see the impact of labeled data on model accuracy. They are introduced to the concept of overfitting, training epochs, and prediction confidence, all within a simple, visual interface. By encouraging active participation e.g., by allowing users to record their own images Teachable Machine fosters experiential learning.\u003c/p\u003e\u003cp\u003eMoreover, the application potential goes beyond image classification. Learners can use the same platform to build models that recognize sound (e.g., voice commands, musical instruments) or body poses (e.g., hand signals, yoga positions). These applications have clear relevance to IoT systems where such inputs might control devices or trigger alerts. The model\u0026rsquo;s ability to be exported and deployed across environments web applications, mobile apps, or embedded devices mirrors the practical deployment of ML in IoT-based projects. These integrations underscore the real-world applicability of concepts learned through Teachable Machine, connecting ML education directly to societal use cases.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study demonstrates that Teachable Machine serves as an effective tool for introducing machine learning to a wide audience. By enabling image classification with only a few examples per class, the platform shows how transfer learning can simplify model training for beginners. Through hands-on experimentation, learners can understand the lifecycle of an ML model from data gathering to deployment without needing to write code. The flexible export options and privacy-preserving design of Teachable Machine further enhance its suitability for education. It can serve as a gateway to more advanced ML tools and, eventually, to integrated systems involving IoT and real-time automation. For institutions and educators looking to integrate AI into their curriculum or outreach, Teachable Machine offers a powerful starting point.\u003c/p\u003e"},{"header":"6. Future Work","content":"\u003cp\u003eWhile this project focused on a basic image classification task with four animal classes, future work can explore additional data modalities and applications. For instance, sound classification could help learners understand how audio features contribute to recognition models. Pose classification could be used in health monitoring or gesture-based device control. Another promising area is the integration of Teachable Machine with physical computing platforms such as Arduino or Raspberry Pi, thereby linking ML models with sensor-driven IoT systems. This would enable learners to simulate real-world smart applications such as a model that uses image data to automate plant disease detection [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], face recognition, home security and automation and many more. Finally, incorporating Teachable Machine into formal education through structured lesson plans and projects aligned with local challenges would promote contextual learning and innovation among youth and grassroots communities.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eContribution\u003c/strong\u003e: Ritu R Lamsal involved in Conceptualization, manuscript drafting, literature review and S Lamsal contributed in data /sample collection, model training and review writing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e: Authors declares no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e: No external funding\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgement:\u003c/strong\u003e The authors would like to thank all the students who encourage us to work on this article. We also extend our sincere Acknowledgement to Google and teachable machine platform.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eLamsal RR, Bhattarai M, Lamsal S (2025) AI and IoT Integration in Engineering: Use Cases, Challenges, and Opportunities in Nepal, Jan. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.13140/RG.2.2.30872.51208\u003c/span\u003e\u003cspan address=\"10.13140/RG.2.2.30872.51208\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eThornton C, Hutter F, Hoos HH, Leyton-Brown K Auto-WEKA: Combined selection and hyperparameter optimization of classification algorithms, in *Proc. 19th ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining*, 2013, pp. 847\u0026ndash;855. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/2487575.2487629\u003c/span\u003e\u003cspan address=\"10.1145/2487575.2487629\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRaschka S, Patterson J, Nolet C Machine learning in Python: Main developments and technology trends in data science, machine learning, and artificial intelligence, *Information*, vol. 11, no. 4, p. 193, 2020. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/info11040193\u003c/span\u003e\u003cspan address=\"10.3390/info11040193\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoogle (2024) Teachable Machine: Train a computer to recognize your own images, sounds, \u0026amp; poses, [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://teachablemachine.withgoogle.com\u003c/span\u003e\u003cspan address=\"https://teachablemachine.withgoogle.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLamsal RR, Karthikeyan P, Otero P, Ariza A (1900) Design and implementation of Internet of Things (IoT) platform targeted for smallholder farmers: From Nepal perspective, *Agriculture*, vol. 13, no. 10, p. 2023. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/agriculture13101900\u003c/span\u003e\u003cspan address=\"10.3390/agriculture13101900\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLamsal RR, Acharya UK, Karthikeyan P, Otero P, Ariza Quintana A (2024) Implementing Internet of Things for real-time monitoring and regulation in off-season grafting and post-harvest storage in citrus cultivation: A case study from the hilly regions of Nepal. *Manuscript submitted for publication*\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR. R. Lamsal *et al*., Monitoring and regulating climatic condition of polyhouse for successful off-season grafting of citrus fruits using Internet of Things platform, *Int. J. Agric. Environ. Biotechnol.*, vol. 15, Special Issue, pp. 417\u0026ndash;422, (2022) [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://aiasa.org.in/wp-content/uploads/2022/09/IJAEB-Sp-Issue-AAFS-2022-1-1.pdf\u003c/span\u003e\u003cspan address=\"https://aiasa.org.in/wp-content/uploads/2022/09/IJAEB-Sp-Issue-AAFS-2022-1-1.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM. A. Almaiah *et al*., Machine learning adoption in educational institutions: Role of Internet of Things and digital educational platforms, *Sustainability*, vol. 15, no. 5, p. 4000, (2023)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eA. G. Howard *et al*., MobileNets: Efficient convolutional neural networks for mobile vision applications, *arXiv preprint*, arXiv:1704.04861, 2017. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/1704.04861\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/1704.04861\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen D, Chen Y, Li Y, Mao F, He Y, Xue H (2021), June Self-supervised learning for few-shot image classification. In \u003cem\u003eICASSP 2021\u0026ndash;2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)\u003c/em\u003e (pp. 1745\u0026ndash;1749). IEEE\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoodfellow I, Bengio Y, Courville A, *Deep Learning*, Cambridge MA (2016) MIT Press, [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://alvarestech.com/temp/deep/Deep%20Learning%20by%20Ian%20Goodfellow,%20Yoshua%20Bengio,%20Aaron%20Courville%20\u003c/span\u003e\u003cspan address=\"http://alvarestech.com/temp/deep/Deep%20Learning%20by%20Ian%20Goodfellow,%20Yoshua%20Bengio,%20Aaron%20Courville%20\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e(z-lib.org).pdf [Accessed: May 10, 2025]\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKrizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural networks, in *Advances in Neural Inf. Process. Syst.*, vol. 25, pp. 1097\u0026ndash;1105\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSimonyan K, Zisserman A Very deep convolutional networks for large-scale image recognition, *arXiv preprint*, arXiv:1409.1556, 2014. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/1409.1556\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/1409.1556\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition, in *Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR)*, pp. 770\u0026ndash;778\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHuang G, Liu Z, van der Maaten L, Weinberger KQ (2017) Densely connected convolutional networks, in *Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR)*, pp. 4700\u0026ndash;4708\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIandola FN, Moskewicz MW, Ashraf K, Golz M, Keutzer K SqueezeNet: AlexNet-level accuracy with 50\u0026times; fewer parameters and \u0026lt;\u0026thinsp;0.5MB model size, *arXiv preprint*, arXiv:1602.07360, 2016. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/1602.07360\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/1602.07360\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan M, Le QV EfficientNet: Rethinking model scaling for convolutional neural networks, in *Proc. Int. Conf. Mach. Learn. (ICML)*, 2019, pp. 6105\u0026ndash;6114. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/1905.11946\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/1905.11946\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan M, Le QV EfficientNetV2: Smaller models and faster training, in *Proc. Int. Conf. Mach. Learn. (ICML)*, 2021. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2104.0029\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2104.0029\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDosovitskiy *et A al*., An image is worth 16\u0026times;16 words: Transformers for image recognition at scale, in *Proc. Int. Conf. Learn. Represent. (ICLR)*, 2020. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2010.11929\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2010.11929\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu *et al* Z et al (2021) Swin Transformer: Hierarchical vision transformer using shifted windows, in *Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV)*, pp. 10012\u0026ndash;10022. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2103.14030\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2103.14030\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu *et al*. Z et al A ConvNet for the 2020s, in *Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR)*, 2022, pp. 11976\u0026ndash;11986. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arxiv.org/abs/2201.03545\u003c/span\u003e\u003cspan address=\"https://arxiv.org/abs/2201.03545\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLamsal RR, Acharya P, Pokhrel R et al Integrating Machine Learning and RAG-Based Chatbot for Mandarin Orange Disease Detection in Hilly Region of Nepal, 27 February 2025, PREPRINT (Version 1) available at Research Square [\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21203/rs.3.rs-6105402/v1]\u003c/span\u003e\u003cspan address=\"10.21203/rs.3.rs-6105402/v1]\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Madan Bhandari University of Science and Technology","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Teachable Machine, Machine Learning for Beginners, Image Classification, IoT and ML, No-Code ML, MobileNet, TensorFlow","lastPublishedDoi":"10.21203/rs.3.rs-7625905/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7625905/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe technical complexity associated with traditional machine learning (ML) tools often impedes early-stage learners and non-programmers from exploring artificial intelligence (AI). This paper presents a pedagogically intuitive approach using Google\u0026rsquo;s Teachable Machine a no-code platform that allows users to create ML models via a graphical interface. We conducted an educational experiment where novice learners built an image classifier to distinguish between animal classes. Our experiment involved the classification of four animal class Cat, Dog, Monkey, and Tiger using a minimal dataset of just 3\u0026ndash;5 images per class to ensure simplicity for beginners. Using transfer learning with MobileNet architecture and basic training settings (50 epochs, 0.001 learning rate), the model achieved high classification accuracy 100% for visually distinct classes and up to 97% for more similar ones. The platform\u0026rsquo;s ability to export trained models to TensorFlow, TensorFlow Lite, expands its potential for real-world deployment. This study concludes that GUI-based ML tools can democratize AI education and discuss their integration with IoT applications, making it accessible, interactive, and impactful for beginners.\u003c/p\u003e","manuscriptTitle":"Introducing Image Classification through No-Code Teachable Machine Platform: An Intuitive Approach for Novice Learners","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-17 11:02:33","doi":"10.21203/rs.3.rs-7625905/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"34b2c033-3a56-47f8-9a0a-824e6e33bedc","owner":[],"postedDate":"September 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":54778649,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2025-09-17T11:02:33+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-17 11:02:33","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7625905","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7625905","identity":"rs-7625905","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}