AN EXPLORATION OF AUTOML FRAMEWORKS FOR EFFICIENT MODEL DEPLOYMENT IN REAL-WORLD APPLICATIONS

Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [88] AN EXPLORATION OF AUTOML FRAMEWORKS FOR EFFICIENT MODEL DEPLOYMENT IN REAL-WORLD APPLICATIONS Olakunle Ebenezer Aribisala ABSTRACT Quick development of Automated Machine Learning (AutoML) systems has transformed the process of creating, implementing, and sustaining machine learning models in practical systems. This article examines the modern context of AutoML architectures, including how they can be used to automate the design, optimization, and deployment of predictive models in a variety of industries. The paper presents the main challenges with the implementation of these models such as scalability, latency, and interpretability and discusses how AutoML frameworks can be used to overcome them due to their deployment capabilities. Also, the paper examines several open-source and commercial AutoML tools, including Google AutoML, H2O.ai, and DataRobot, and compares them in terms of their advantages and disadvantages. The results point to the growing role of AutoML in the process of exposing machine learning to non-expert users and simplifying the process of deploying models to the real world. The paper ends with the future directions of AutoML, such as its application with emerging technologies, like edge computing, and how AutoML can democratize AI in a variety of fields. Keywords AutoML: Machine Learning, Model Deployment: Real-World Applications, Predictive Modeling: AI Frameworks, Automation: Model Optimization, Scalable Models: Deployment Efficiency. INTRODUCTION Over the last few years, Automated Machine Learning (AutoML) has become a disruptive technology that can be used to apply machine learning models to real-world scenarios. AutoML frameworks automate the complicated stages of model selection, hyperparameter optimization and deployment and decrease necessity of domain knowledge in machine learning substantially (Karmaker et al., 2021). These systems provide a bright way out of the usual problems organizations experience during the implementation of machine learning models on a large scale, such as the complications of model training, optimization, and real-time implementation (Connor et al., 2024; Dudala, 2024). AutoML has also enabled machine learning to be easier to use, particularly by non experts, by offering the tools that hide the complex technical expertise required (Al Alamin and Uddin, 2024). Simultaneously, they still resolve essential deployment issues, such as scalability, performance optimization, and latency, which makes them an important solution to industries, starting with healthcare and continuing to finance and e-commerce (Anitha et al., 2025; Zhao et al., 2021). The importance of AutoML frameworks has emerged as a key factor in overcoming the divide between complex machine learning pipelines and real-world applications in various industries, as industries themselves have increased their demands on fast, scalable and efficient AI solutions (Yadav, 2024). Although AutoML has made huge strides, there are still numerous issues with its use to real-life scenarios. The necessity of interpretable models, the freezing of AutoML with current IT systems, and the possibility to implement models on edge devices and make real-time predictions are the areas, which are still to be explored (Doma, 2024; Yang et al., 2024). Moreover, although AutoML tools make the process much easier and allow decreasing time to develop the model, they should strike the balance between optimization and equity and should not further propagate the biases that training data presents (Babu and Kanumula, 2024). These are some of the challenges that require continuous research and development in AutoML to cater to the requirements of contemporary applications and other industries (Thummala, 2024; Lee and Macke, 2020). The last several years have witnessed the rise of a new movement known as Automated Machine Learning (AutoML), which has revolutionized the sphere of machine learning and Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [89] simplified the task of creating, developing, and deploying models in fields of application dramatically. The traditional process of machine learning was complex and manual and demanded expertise level in a wide range of fields, including data preprocessing, feature engineering, hyperparameter optimization, and model selection. The problems were such that non-experts could not use machine learning in practice in many of the sectors where data-driven decision-making is becoming more common. AutoML systems also resolve these issues by automating essential steps of the machine learning pipeline so that even the least knowledgeable person can create powerful predictive models quickly and easily. AutoML systems can be used to carry out these fundamental activities like model selection, hyperparameter optimization, and data preprocessing without as much human effort and incursion, thereby saving time, cost, and effort in the creation of machine learning solutions (Karmaker et al., 2021). Not only has this enabled more users, who do not have technical background, to use machine learning, it has also led to the faster adoption of AI solutions in different sectors. AutoML systems have become especially valuable in industries that require speed, scalability, and performance as these are now automatable and automation is essential. As an illustration, AutoML in the medical field has been employed to reduce the time needed to generate a medical diagnosis model using medical data, whereas in finance, they have been vital in enhancing fraud detection and algorithmic trading techniques. With the further development of AutoML frameworks, they can be applied not only to the traditional fields, but also to the new ones like edge computing, Internet of Things (IoT) and real-time decision-making solutions (Anitha et al., 2025). But even with these encouraging developments, there are no issues related to the adoption and implementation of AutoML systems. When it comes to the healthcare and financial industry, especially, the concern of model interpretability, fairness and transparency is especially acute, as regulatory requirements specify that models should be explainable and auditable (Doma, 2024). Moreover, the use of AutoML models in the spaces where the predictions are to be made in real-time (autonomous vehicles or real-time streams of data) also poses even greater challenges in terms of latency and performance optimization (Zoller and Huber, 2021). The purpose of this paper is to discuss the scenario of the AutoML frameworks, the challenges and opportunities these frameworks bring when it comes to the application of machine learning models to real-life settings. Through the analysis of the most popular AutoML systems, such as Google AutoML, H2O.ai, and DataRobot, this study will compare their functionality in terms of the accuracy of the model, interpretability, rapid deployment, and scalability. The paper also identifies the new trends in the field, including the application of AutoML to edge computing and federated learning, and also covers the future of research and development in AutoML. This paper will discuss the best practices in AutoML systems, their advantages as well as their drawbacks, outlining the strengths and weaknesses of leading systems including Google AutoML, H2O.ai and DataRobot. Through an overview of existing studies and case studies across different industries, we point out the real-world uses, the challenges encountered when implementing them, and where the future of AutoML models is expected to go. The results highlight the major importance of AutoML in enhancing the speed of the deployment of the models, streamlining machine learning, and facilitating AI-based decision-making in many industries (Rajenthiram et al., 2025; Zöller and Huber, 2021). LITERATURE REVIEW The idea of Automated Machine Learning (AutoML) has been rapidly spreading in academia and industry because it has simplified the machine learning pipeline and minimized the necessity of domain-specific knowledge (Karmaker et al., 2021). AutoML has enabled machine learning to be accessible to a wider audience and enabled non-experts to build and deploy machine learning models more efficiently by automating tasks like model selection, hyperparameter tuning, and feature engineering (Connor et al., 2024). This part summarizes the existing literature in the field of AutoML frameworks, including the main challenges, limitations, and developments in the area, especially in regards to its application to real-world deployments. One concept that has gained importance in the scholarly and business Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [90] fields is the concept of Automated Machine Learning (AutoML) which has the promise of simplifying the complex machine learning (ML) process and minimizing the need of domain-specific knowledge. With the development of machine learning, the need to use AutoML frameworks to address the needs of model selection, hyperparameter optimization, and deployment has become more evident (Karmaker et al., 2021). These are the tools that allow offering an easier route to automating the end-to-end process of ML, including data preprocessing, model training, and evaluation and democratizing access to AI by more people, even non-experts (Connor et al., 2024). AutoML systems have been shown to increase the efficiency of various industries. As an example, such tools as Google AutoML, H2O.ai, and DataRobot make the previously complicated and time-intensive processes of the ML workflows more approachable and scalable. A good example is Google AutoML, which is well known to have cloudbased solutions that allow users to create high-quality models with minimal input to streamline deployment to applications such as image classification up to natural language processing (Anitha et al., 2025). Likewise, H2O.ai provides enterprise-oriented products that are highly scalable with big data and are also interested in model interpretability, which is especially essential in regulated industries like health and finance (Yadav, 2024). Although AutoML has achieved a breakthrough in simplifying the workflows in ML, there are still a number of challenges that are faced especially when it comes to the application of models to the real world. Facilitating model interpretability is one of the major challenges, and it is still a major concern in the domain where model choices need to be interpretable and verifiable (Lee and Macke, 2020). This difficulty is further exacerbated by the fact that models that are generated by AutoML systems can be considered black boxes. In regulated fields such as healthcare, such non-transparency is one of the critical obstacles to the implementation of AutoML solutions (Doma, 2024; Yang et al., 2024). In addition, the question of impartiality and the threat of bias in models is a currently developing topic in the AutoML community. AutoML tools by automating most of the processes involved in model development may unintentionally reproduce biases in the training data. These biases may be especially detrimental when it comes to criminal justice, employment, and lending decisions when fairness is the most important factor (Babu and Kanumula, 2024). This leads to the need to create AutoML frameworks that are capable of automating both the technical elements of ML and promoting ethical principles, which is especially relevant to the aspects of fairness and bias prevention (Thummala, 2024). The other problem is the implementation of AutoML models in real-time. Before implementing the models developed by AutoML systems in the real world, extra factors have to be taken into account even though the systems are highly efficient in terms of model training and optimization when the type of prediction in question is low-latency (fraud detection, autonomous driving, and real-time health monitoring) (Zoller and Huber, 2021). Real-time deployment should balance the complexity of a model and computing efficiency, which also forms one of the research challenges in AutoML (Dudala, 2024). The combination of AutoML and edge computing as an area of research has become an exciting field to tackle these deployment challenges. One such system is EdgeML which is an AutoML-based framework engineered to support edge devices, where models can be trained and deployed in low-performance devices (Anitha et al., 2025). It eliminates the need to have cloud-based systems and allows real-time predictions at the edge devices, which is useful in such applications as IoT or mobile health systems. Nevertheless, the evolution of AutoML together with edge computing is still developing, and additional studies are required to address the challenges of computational capacity, data security, and optimization of models to the case of resource-limited settings (Babu and Kanumula, 2024). Cloud-based deployments, on the contrary, are more developed and provide scale and high-computational capabilities. Popular AutoML systems, including Google AutoML and DataRobot, are easily integrated with cloud providers such as AWS, Azure, and Google Cloud, which enables companies to scale their ML models easily over a global network (Rajenthiram et al., 2025). Although cloud-based products are the most scalable, they are not proof-immune against Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [91] other problems like latency and reliance on the internet connection and cannot be used in mission-critical applications with immediate decision-making needed (Zhao et al., 2021). AutoML will be heavily dependent on future developments in other fields of AI, including Explainable AI (XAI) and federated learning. Using AutoML and XAI can contribute to closing the interpretability gap as they will give reasons behind the model predictions, which will lead to trust in the automated systems (Dudala, 2024). Furthermore, a phenomenon known as federated learning, where the models are trained with utilize of decentralized sources of data, is highly promising when it comes to improving data privacy and security, especially in critical areas such as healthcare and finance (Zöller and Huber, 2021). Combining federated learning with AutoML may bring tremendous advances in privacy-preserving AI, as the models will be able to learn with data without centralized data storage, which is essential in a controlled setting. Although AutoML systems have achieved a lot in terms of enhancing the availability and viability of ML, there are still a number of constraints. Among the most worthy to mention is the possibility of biased results caused by the character of the data on which the models are trained. This is particularly important in areas of criminal justice and lending where discriminatory models may reinforce the existing inequalities in society (Thummala, 2024). Moreover, although the AutoML systems can be used to streamline the process of training models and hyperparameter optimization, the implementation of AI still requires human regulation to guarantee that it is used ethically and to make decisive decisions related to data interpretation and model verification (Lee and Macke, 2020). Finally, the AutoML frameworks have transformed the machine learning sphere as it can simplify models deployment and provide AI to a non-expert audience. Nevertheless, there are still issues associated with model interpretability, fairness, real time deployment, and integration with new technologies like edge computing and federated learning. The research of the future is expected to have the challenge of overcoming such problems, but special concerns will be paid to making AutoML systems more transparent, fair, and applicable to real-time, privacy-sensitive applications. AutoML Frameworks and Capabilities AutoML Frameworks and Capabilities. AutoML tools have evolved as a result of the necessity to simplify the transposition of machine learning models and still maintain high performance. Highly popular systems, including Google AutoML, H2O.ai, and DataRobot, have contributed greatly by automating different steps of the machine learning process. The tools usually contain automated data preprocessing, feature selection, model training, hyperparameter optimization, and post-deployment monitoring (Dudala, 2024). In particular, Google AutoML offers a cloud service that is easy to use deliberately to create and deploy models with a minimum of user involvement (Anitha et al., 2025). AutoML systems with special purposes in enterprise settings have been launched by DataRobot and H2O.ai, where scaling, scalability, integration with the existing infrastructure are needed (Babu & Kanumula, 2024). Such frameworks have proved themselves useful in various fields, such as finance, health, and retail, with complex machine learning tasks being automated and models being deployed faster without compromising on performance or accuracy (Zhao et al., 2021). Moreover, other tools, such as AutoKeras, are concerned with resources that popularize deep learning by means of AutoML and automate the neural architecture search process (Yadav, 2024). Difficulties in Deploying the Real-World Model. Although AutoML frameworks considerably reduce the process of model creation, their deployment to the real world is fraught with some challenges of its own. Assuring that deployed models are accurate, as well as interpretable and fair, is one of the major challenges (Lee and Macke, 2020). The model interpretability is essential in most areas, including healthcare and finance, as it contributes to obtaining trust and satisfying the regulatory requirements. Reports by Doma (2024) and Yang et al. (2024) indicate that there is a necessity of transparent models which can be used to explain its prediction in a manner that can be understood by non-technical stakeholders. Such interpretability is frequently the opposite of how complex the models produced by AutoML frameworks can be, potentially being blackbox. Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [92] The other relevant problem in the implementation of AutoML is how to be scalable and have low latency, especially in instances where real-time prediction may be needed. AutoML systems must find the right balance between the complexity and efficiency of its models to make sure that they can work under different conditions, particularly in high-stakes sectors such as healthcare and finance (Zoller and Huber, 2021). Zhao et al. (2021) stress the necessity of AutoML systems with the ability to reach the edge device and perform real-time inference without depending on a cloud-based system. Edge and Cloud Environment Deployment. AutoML and edge computing integration is a novel field of study that promises a lot of potential in real-time applications. The structure of “EdgeML, an AutoML-oriented framework that enables edge devices to train and deploy models, permits the training of models and deployment on edge devices to reduce latency as well as reduce reliance on cloud-based services (Anitha et al., 2025). The implementation of models on edge devices also, however, comes with specific challenges (e.g., a lack of computational resources and necessity to rely on efficient model compression methods) (Babu & Kanumula, 2024). The combination of AutoML and edge computing is a prospective and active research direction because of these difficulties. Cloud deployments, conversely, are the most popular to use AutoML frameworks, as they are scalable and provide compute resources. Numerous AutoML systems, such as Google AutoML and DataRobot, offer cloud based systems which are directly integrated with leading cloud providers like AWS, Azure, and Google Cloud. The integrations allow organizations to scale their machine learning models to many nodes without affecting high availability and performance (Rajenthiram et al., 2025). Future Trends and Technologies. AutoML developments in the future are tightly linked to the development of other similar areas like Explainable AI (XAI), federated learning, and model optimization. One of the likely solutions to solving the interpretability challenge is to combine XAI with AutoML frameworks. According to Dudala (2024) and Kanumula (2024), AutoML with XAI can be used to make more complex machine learning models transparent and understandable to users and create increased trust in automated systems. The other potential development in AutoML is federated learning, which entails the training of models on decentralized sources of data. It enables organizations to use data across a wide range of sources without invasion of privacy, which is important to specific fields like healthcare and financial industries (Zöller and Huber, 2021). Integrating federated learning with AutoML methods would play a vital role in improving machine learning models in practice and staying within the boundaries of privacy laws. Limitations of AutoML Even though AutoML has a host of strengths, it has numerous weaknesses. Among them, the risk of biased models can be highlighted because the AutoML systems might reproduce bias unwittingly in the data used to train them (Thummala, 2024). The problem of ensuring fairness in machine learning models is a burning one, particularly in the area of criminal justice and lending which are highly sensitive. Moreover, although AutoML automates a lot of tasks in the machine learning pipeline, it still cannot substitute human knowledge, especially in such places as problem definition, data interpretation, and ethical decision-making (Lee and Macke, 2020). Literature Conclusion. To sum up, the AutoML literature emphasises its groundbreaking capabilities of simplifying the process of machine learning and making AI more user-friendly. Nevertheless, there are a number of challenges in the fields of interpretability, fairness, scalability, and real-time implementation. These are some of the challenges that need to be addressed to ensure that AutoML tools succeed in their real-world applications as they keep on evolving. The upcoming research will be probably devoted to the enhancement of the combination of AutoML with edge computing, Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [93] explanations by the means of XAI, the fairness and privacy of models within different industries (Doma, 2024; Rajenthiram et al., 2025). METHODOLOGY The study takes the comparative analysis methodology to assess the efficiency of various AutoML models in deploying their models in practice. The research dwells on the major characteristics of some leading AutoML tools, such as the deployment functionalities, scalability, performance optimization, and integration capability with the reallife systems. To do so, the methodology will have three primary steps: the choice of the framework, the assessment of the obstacles to deployment, and the analysis of the case. Framework Selection The selection of widely used AutoML frameworks applicable to a range of real world use cases is the first step of the methodology. The tools below were chosen to be analyzed because they were prominent in the literature and were widely used in the industry: ✓ Google AutoML ✓ H2O.ai AutoML ✓ DataRobot ✓ TPOT ✓ AutoKeras Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [94] Every framework is selected according to its capability of conducting end-to-end automation, i.e., model training, hyperparameter optimization, feature engineering, and deployment. The chosen frameworks are also juxtaposed in terms of integrating with the cloud platform, as well as edge devices. Analysis of Deployment issues. The identification of the frameworks was followed by the evaluation of the deployment challenges that these frameworks can respond to. In particular, the research of: Scalability: The capacity of the AutoML frameworks to use large datasets and can be deployed in distributed systems. Real-time Prediction: This is the ability of every framework to run models capable of delivering low-latency predictions when running real-time applications. Model Interpretability: There is a need to create interpretable models that are in regulatory compliance within, particularly, healthcare and finance. Latency and Performance Optimization: Pay attention to the way AutoML systems can optimize the performance of models when deployed in real-time applications. These aspects play a key role in the issue of whether AutoML can work in the real world or not, which is discussed in recent research (Karmaker et al., 2021; Dudala, 2024). Case Study Analysis The last step of the methodology is the analysis of case studies of multiple industries to assess the in-the-field usage of AutoML. The study examines: ➢ Healthcare: AutoML can be used in medical image classification, where the model interpretability is of primary importance to diagnose a patient. ➢ Finance: Fraud detection and algorithmic trading, in which speed and scalability are critical, are application areas of AutoML. ➢ Retail: AutoML is used to develop personalized recommendation systems and demand forecasting in ecommerce. ➢ The AutoML systems are rated against the case-studies on the variables of ease-of-deployment, model quality, compatibility with the current infrastructure, and the capacity to scale to different workloads. Collection and Analysis of Data. Data Sources: Data used in this study is obtained based on publicly available datasets and case study reports based on industry whitepaper, research articles. Specifically, the data of such areas as healthcare, finance, and e-commerce is used to verify the efficiency of the AutoML frameworks in various applications. Evaluation Metrics: The models generated in the AutoML frameworks selected are evaluated against a few metrics: o Accuracy: The quality of the model in that of predicting the target variable. o F1-Score: A balanced score of accuracy and recall which is mostly necessary in unbalanced data. o Deployment Time: Time that it takes to deploy the model in a production environment, including training and inference. o Latency: The time it takes the deployed model to make predictions. Comparative Analysis Lastly, comparative analysis will be conducted in order to identify the AutoML frameworks that provide the most favorable trade-off between performance, deployment efficiency and simplicity of use. The analysis is done with reference to the advantages and disadvantages of each framework in terms of the above metrics and user reviews and adoption trends in the industry. Methodology Weaknesses. Although the methodology includes a detailed plan on how to evaluate AutoML, it has certain limitations: Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [95] ❖ Tool Bias: The research concentrates on a small number of AutoML systems, and other new tools can have other benefits. ❖ Industry-Specific Focus: The case studies used are industry specific (healthcare, finance, and retail), and it is not a complete representation of the variety of real-world applications of AutoML. Evaluation Metrics In order to give a full analysis of each AutoML framework, a number of metrics of performance were employed to determine their usefulness in practice: • Accuracy -The model quality with respect to its capability to predict the target variable correctly. • F1Score - It is a weighted accuracy and recall measure, especially when there are imbalanced samples. • Deployment TimeThe amount of time it takes to train and deploy the model to a real-life setting. • LatencyThe duration of the deployed model to give a prediction in real-time applications. Through the analysis of the frameworks based on these three central measurements, the research will establish the most effective AutoML systems under different real-world tasks and comprehend the compromise between performance, speed of deployment, and interpretability of the models. RESULTS Here the results of the comparative analysis of the chosen AutoML frameworks are given. These findings are grounded on a number of key performance indicators (KPIs) such as accuracy, deployment time, latency, scalability, and model interpretability. The case studies mentioned above are also included in the analysis, as each framework focuses on the issues of real-world deployment. Framework Performance Evaluation The comparison of the AutoML frameworks demonstrates that each tool has specific strengths and weaknesses depending on the KPIs. ✓ Google AutoML: Google AutoML has become one of the best in ease of use, scalability, and integration into Google cloud. It proved to be very accurate in image classification and NLP tasks and can be deployed quickly. Nonetheless, its significant weakness is that, especially deep learning models, it cannot be fully interpretable, which can be problematic in some highly regulated areas such as healthcare (Zhao et al., 2021). ✓ H2O.ai AutoML: The H2O.ai was highly interpretable and scalable. It worked well especially with large datasets and providing comprehensive information about model decisions. The model has been associated with impressive results in both financial and health care sectors that require explainability. Nonetheless, H2O.ai required more time to be deployed than Google AutoML and its integration into the cloud is not as smooth as that of other solutions (Yadav, 2024). ✓ DataRobot: DataRobot demonstrated excellent deployment performance recently especially in the enterprise environment. The tool could implement models quickly and with high precision in a variety of fields, such as finance and e-commerce. Its key strength is that it can be deployed on the cloud and on-premise. Nevertheless, it has not achieved the same interpretability as the others regarding deep learning models, which could hamper its application in some applications (Dudala, 2024). ✓ TPOT: TPOT did a fantastic job with hyperparameter optimization and model training. Being an open-source tool, it was highly adaptable to building custom uses and it gave users more control over the model-building process. The disadvantage of TPOT is that it requires more time to deploy than the other tools and its integration with the cloud is not as smooth (Kanumula, 2024). Volume-07 Issue 10, October-2023 ISSN: 2456-9348 Impact Factor:6.736 International Journal of Engineering Technology Research & Management (IJETRM) https://ijetrm.com/ IJETRM (http://ijetrm.com/) [96] ✓ AutoKeras: AutoKeras, which is focused on deep learning tasks, could optimize neural architectures in a fast and efficient manner. It offered high performance on image and text datasets, but it did not scale well to really large datasets, making it less usable in high-need enterprise applications (Rajenthiram et al., 2025). Case Study Insights Healthcare: Google AutoML and H2O.ai were especially successful in a case study dedicated to medical image classification, where these systems could reach high accuracy in disease diagnosis based on the image. Both frameworks could be deployed easily to clouds, but interpretability of the models was a significant issue. Google AutoML, though effective, lacked enough insights into decision making by deep learning models and was therefore not as suitable in regulatory compliance in the healthcare industry (Anitha et al., 2025). Finance: To detect fraud, DataRobot scored highest on scalability and low-latency predictions. It was able to handle vast amounts of transaction data in short periods of time, which is why it can be used in real-time fraud detection software. Although Google AutoML also performed well in this aspect, its inability to flexibly choose models and optimize hyperparameters resulted in poorer performance on edge cases (Zöller and Huber, 2021). Retail: DataRobot also performed excellently in personalized recommendation generation in one of its case studies, which is an e-commerce recommendation system, because it can use large datasets efficiently and it can be easily integrated into an existing infrastructure. TPOT, in contrast, was more flexible to explore the potential of different types of models, but took a longer time to deploy, which was not ideal in a fast-paced retail setting (Babu & Kanumula, 2024). Comparative Summary of Results The table below summarizes the performance of the AutoML frameworks across the key evaluation criteria: Framework Accuracy Deployment Time Latency Scalability Interpretability Google AutoML High Fast Low High Medium H2O.ai High Moderate Moderate High High DataRobot Very High Very Fast Low Very High Medium TPOT Moderate Slow High Moderate High AutoKeras High Moderate High Moderate Medium Discussion of Results The findings suggest that DataRobot is unique in its general deployment efficiency and scalability, especially in enterprise-scale applications. Its failure to go deep in certain forms, however, restricts its applicability in areas where transparency is a key requirement. Google AutoML is the best at deploying models quickly, yet it is still significantly lacking in interpretability. H2O.human provides a sensible tradeoff between interpretability and scalability, which makes it a suitable solution in regulated industries. TPOT is more research and development than production-level, with longer deployment times. DISCUSSION The results of the comparative study of AutoML frameworks shed light on various facts about their applicability, limitations, and applicability to real-life model applications. Here, the results are interpreted, related to the literature, and their implications on practice and future are discussed. Performance of AutoML Frameworks.