From challenges to implementation and acceptance: Addressing key barriers in artificial intelligence, machine learning, and deep learning
Synopsis
Machine learning (ML) and deep learning (DL) have transformed different industries by facilitating sophisticated data analysis, predictive modeling, and autonomous decision-making. Despite the ability to greatly change things, there are many obstacles preventing their widespread use and impact. A major obstacle is the challenge of data quality and quantity; ML and DL models need large amounts of high-quality, labeled data, which can be hard and expensive to acquire. Moreover, the innate intricacy of these models frequently results in a dearth of clarity and visibility, posing difficulties in comprehending and having faith in their decision-making procedures. This has caused worries about ethical ramifications and favoritism, since models may unknowingly continue current biases found in the data used for training. Moreover, the fast rate of technological progress leads to a constantly changing environment, requiring practitioners and organizations to continuously learn and adapt. Security and privacy concerns are significant challenges due to the susceptibility of ML and DL models to attacks and breaches, jeopardizing the security of private data. Additionally, incorporating ML and DL into current systems and processes presents challenges such as requiring unique knowledge and ensuring that technological solutions align with business goals.
Keywords: Artificial intelligence, Deep learning, Machine learning, Challenges, Data Quality, Data Privacy and Security.
Citation: Rane, N. L., Mallick, S. K., Kaya, O., & Rane, J. (2024). From challenges to implementation and acceptance: Addressing key barriers in artificial intelligence, machine learning, and deep learning. In Applied Machine Learning and Deep Learning: Architectures and Techniques (pp. 153-166). Deep Science Publishing. https://doi.org/10.70593/978-81-981271-4-3_8
8.1 Introduction
Machine learning (ML) and deep learning (DL) have transformed many industries with their ability to offer advanced data analysis, pattern recognition, and predictive features (Thakkar & Lohiya, 2021; Kocher & Kumar, 2021; Wani et al., 2022). These technologies have enabled substantial progress in fields like healthcare, finance, autonomous systems, and natural language processing (Rani et al., 2022; Bhatt et al., 2021; Goshisht, 2024). Although they are widely utilized and achieve success, ML and DL face obstacles that prevent them from reaching their maximum capabilities. These challenges vary from technical problems like model interpretability and data quality to ethical issues such as bias and privacy concerns (Kocher & Kumar, 2021; Wani et al., 2022; Rani et al., 2022; Mohammed & Kora, 2023; Nasir & Sassani, 2021). A major obstacle in ML and DL is understanding complex models, particularly deep neural networks (Wani et al., 2022; Shamshirband et al., 2021; Al-amri et al., 2021; Bhushan et al., 2023). As these models advance, it becomes harder to comprehend and clarify how they make decisions (Wani et al., 2022; Jdey et al., 2022; Alzubaidi et al., 2021; Alafif et al., 2021). This lack of transparency can pose a significant challenge in critical applications where being open and responsible is crucial. Furthermore, how well ML and DL models perform greatly relies on the quality and quantity of data (Kocher & Kumar, 2021; Talaei Khoei et al., 2023; Chiche & Yitagesu, 2022; Fikri et al., 2021). Inaccurate predictions and unreliable outcomes may result from poor data quality, including incomplete, noisy, or biased datasets. Furthermore, the computational demands for training expansive DL models can be excessively high, requiring a substantial investment in hardware and energy usage. Table 8.1 summarizes the various application areas of machine learning and deep learning, the methods used in these areas, success metrics, and challenges faced. A wide range of applications such as image recognition, natural language processing, anomaly detection, recommendation systems, voice recognition, robotic control, financial forecasting and health analytic have been evaluated with their own specific methods and metrics (L’heureux et al., 2017; Kaluarachchi et al., 2021; Khademi et al., 2023). In addition, the unique challenges of each application area are stated and ways to overcome these challenges are discussed. This table provides a comprehensive perspective to understand how machine learning and deep learning techniques are used in industry and what challenges they face.
Table 8.1 Application Areas, Success Metrics and Challenges of ML and DL
Application area
Methods
Success Metrics
Challenges
Explanation
Image Recognition
CNN, Transfer Learning
Accuracy, F1 Score, Kappa Statistics
Cost of data labeling, High processing power requirement
Medical image analysis, face detection
NLP
RNN, Transformers
Accuracy, Bleu Score, Perplexity
Language diversity, ambiguity
Text classification, machine translation
Anomaly Detection
Autoencoders, Isolation Forest
Accuracy, F1 Score, ROC-AUC
False positive rate
Fraud detection, network security
Recommendation Systems
Collaborative Filtering, Matrix Factorization
RMSE, MAE, Precision
Scalability
Movie and product recommendations
Voice Recognition
Hidden Markov Models (HMM), Deep Neural Networks (DNN)
Accuracy, Word Error Rate (WER)
Noise resistance, Variety of accents
Speech recognition, voice assistants
Robotic Control
Reinforcement Learning, Q-Learning
Average Reward, Success Rate
Real-time computing, Security and reliability
Autonomous vehicles, robotic arms
Financial Prediction
Time Series Analysis, LSTM
MAPE, RMSE, R^2
Market fluctuations, Data biases
Stock forecast, risk analysis
Health Analytic
Random Forest, Support Vector Machines (SVM)
Accuracy, AUC, Sensitivity, Specificity
Data privacy, Data heterogeneity
Disease diagnosis, patient results analysis
In addition to technical challenges, ML and DL also encounter significant ethical and societal issues (Kocher & Kumar, 2021; Wani et al., 2022; Silva & Najafirad, 2020; Asharf et al., 2020; Ahmed et al., 2023). Bias within algorithms and data sets can continue and exacerbate current disparities, resulting in unjust and prejudicial results (Hong et al., 2020; Fregoso-Aparicio et al., 2021; Dinsdale et al., 2022). Another crucial issue to address in ML applications is maintaining privacy and security, as the utilization of personal and sensitive information raises concerns regarding the protection of data and user agreement (Goshisht, 2024; Abdar et al., 2021; Yuan et al., 2020; Sapoval et al., 2022). Moreover, the fast progress of ML and DL technologies has exceeded the creation of regulatory frameworks, causing a lack in governance and ethical standards (Wani et al., 2022; Mishra et al., 2021; Yuan & Wu, 2021; Li et al., 2021). In this study, we carry out an extensive review of literature to pinpoint and examine the main obstacles in ML and DL. We have made three significant additions to the current knowledge base:
- We conduct a thorough analysis of the technical obstacles in ML and DL, emphasizing model interpretability, data quality, and computational requirements.
- We examine the ethical and social consequences of ML and DL, focusing on concerns regarding bias, privacy, and regulations.
- We suggest possible solutions and directions for future research to tackle these challenges, with the goal of improving the efficiency, equity, and transparency of ML and DL applications.
8.2 Methodology
Firstly, an in-depth search of pertinent academic databases was carried out, such as IEEE Xplore, ACM Digital Library, Google Scholar, and ScienceDirect. The literature search included terms like "barriers in machine learning," "difficulties in deep learning," "limitations of ML," "issues in DL," and "obstacles in artificial intelligence." The exploration concentrated on peer-reviewed articles, conference papers, and respected technical reports released in the past ten years to ensure the incorporation of the latest developments and conversations. Furthermore, strict criteria for inclusion and exclusion were used to sift through the search results. Research will be considered if it focuses on difficulties with machine learning and deep learning, presents either data-driven findings or theoretical insights, and is written in the English language. Research that only looks into the uses of something, without addressing the difficulties, sources that haven't been peer-reviewed, and materials published before 2014 were not considered in order to uphold the quality and importance of the analysis. Next, the chosen articles underwent an in-depth content analysis. This required extracting important details on the different obstacles found, classifying them, and combining the results into cohesive themes. The difficulties were grouped into four main areas: technical, methodological, ethical, and practical. Technical obstacles include problems such as over-fitting, computational complexity, and data quality. Challenges in methodology involve understanding models, requiring extensive data, and ensuring algorithm reliability. Ethical dilemmas focus on issues like prejudice, equity, and openness in machine learning and deep learning models, whereas practical dilemmas involve implementing, scaling, and up-keeping these technologies in practical situations.
8.3 Results and discussions
Challenges and potential solutions in machine learning and deep learning
ML and DL have transformed many areas by allowing systems to learn from data and make decisions with slight human involvement (Kocher & Kumar, 2021; Wani et al., 2022; Rani et al., 2022; Bhatt et al., 2021; Mohammed & Kora, 2023; Nasir & Sassani, 2021; Kaluarachchi et al., 2021; Yeole, 2024; Xu et al., 2021; Taye, 2023). In spite of their notable progress and extensive use, these technologies encounter various inherent difficulties that may impede their effectiveness and acceptance (Goshisht, 2024; Mohammed & Kora, 2023; Nasir & Sassani, 2021; Shamshirband et al., 2021; Rezaeianjouybari & Shang, 2020; Shafay et al., 2023; Liu et al., 2020; Dar et al., 2022).
Fig 8.1 Challenges in DL and ML
Fig 8.1 highlights different learning methods, such as deep learning, online learning, local learning, transfer learning, lifelong learning, and community learning, and the challenges each faces. Each learning paradigm faces unique challenges. While dealing with problems such as deep learning, feature engineering, nonlinearity and data heterogeneity; online learning and local learning face similar challenges such as processing performance, the curse of modularity, class imbalance, and data locality. While transfer learning combats class imbalance and data heterogeneity; lifelong learning faces additional challenges such as data availability and real-time processing. Ensemble learning, on the other hand, involves challenges such as processing performance, the curse of modularity, and concept drift.
Data Quality and Quantity
One of the biggest problems in ML and DL is the lack of diverse and high-quality datasets. This models demand extensive amounts of labeled data for efficient training. Obtaining this data can often be both costly and require a significant amount of time. Furthermore, the importance of data quality cannot be overstated, as low-quality data can result in imprecise models and untrustworthy results. Challenges like incomplete data, interference, and prejudices within datasets can greatly affect the effectiveness of ML models. Ensuring a variety of data is important as models trained on similar data may struggle to perform well in new or real-life situations. Various methods can be utilized to address the problem of both the quality and quantity of data. Methods like creating artificial data and transferring knowledge from one dataset to another can boost the variety and size of datasets. Furthermore, tapping into crowdsourcing and utilizing public datasets can offer a greater amount of labeled data. Developing strong data preprocessing pipelines to cleanse and standardize data can enhance its quality. Active learning involves models asking human annotators for labels on the most informative samples in an iterative process, aiding in the efficient collection of high-quality labeled data.
Data Privacy and Security
Ensuring data privacy and security has become a critical concern due to the rising utilization of ML and DL in sensitive areas such as healthcare, finance, and personal data. It is difficult to protect user data from breaches and unauthorized access, particularly due to the increase in sophisticated cyber-attacks. Furthermore, there is ongoing exploration of methods such as differential privacy and federated learning to train models while protecting individual privacy, although these approaches need additional development before they can be fully effective and widely accepted. In order to safeguard data privacy and security, methods such as differential privacy can be utilized to prevent individual data points from being inferred from aggregated data. Federated learning enables models to be trained on various devices or servers that store local data samples, without sharing them, thus safeguarding privacy. Advanced cryptographic techniques like secure multi-party computation and homomorphic encryption allow for computations to be performed on encrypted data without requiring decryption, maintaining data privacy during the entire process.
Model Interpretability and Explainability
DL models, especially neural networks, are frequently referred to as "black boxes" because of their intricate and opaque nature. The absence of interpretability presents major obstacles, particularly in industries like healthcare, legal, and financial services where comprehending the decision-making process is vital. Creating ways to improve the interpretability and explainability of these models is crucial for establishing trust and guaranteeing ethical implementation. Approaches such as Explainable AI (XAI) are designed to offer understanding on how models reach decisions, however striking a balance between model intricacy and interpretability is still a persistent obstacle. Enhancing the interpretability and explainability of models can be accomplished by creating techniques for Explainable Artificial Intelligence (XAI). Approaches like SHAP, LIME, and attention mechanisms in neural networks offer understanding into how models make decisions. In addition, creating simpler models that are naturally easier to understand, or incorporating a mix of interpretable and complex components in hybrid models, can assist in achieving a balance between performance and explainability.
Fig. 8.2 Sankey diagram on challenges and potential solutions in ML and DL
Computational and Resource Constraints
Training DL models necessitates significant computational resources, such as strong GPUs and vast memory capacity. The significant expense of computation may pose a barrier for numerous organizations, especially those operating on a tight budget. Moreover, the significant energy usage linked to the training of large models is causing worries about its environmental effects. Efficient algorithms and hardware accelerators are being created to tackle these challenges, yet the quick increase in model complexity is still surpassing these advancements. In order to deal with limitations in computing power and resources, methods like pruning, quantization, and knowledge distillation can be used to decrease the size and complexity of models while still maintaining high performance. Leveraging specialized hardware accelerators such as GPUs, TPUs, and custom AI chips can improve computational efficiency. Moreover, cloud-based alternatives and distributed computing platforms such as Apache Spark and TensorFlow can assist in expanding computational resources as needed.
Generalization and Over-fitting
Generalization is when an ML model can effectively handle data it has not seen before. Over-fitting happens when a model becomes too focused on the training data, including noise and outliers, leading to decreased accuracy on test data. Finding the perfect equilibrium between inadequate fitting and excessive fitting is an ongoing struggle in the fields of ML and DL. Methods such as regularization, cross-validation, and dropout are frequently utilized to address over-fitting, though they are not infallible and often necessitate precise adjustments. Several techniques can be utilized to enhance generalization and reduce over-fitting. Methods like L1 and L2 regularization, dropout, and early stopping are useful in avoiding over-fitting. Cross-validation techniques can be employed to guarantee that models perform effectively with unseen data. Data augmentation, the process of generating additional training examples by modifying current data, can improve generalization as well. Furthermore, ensemble learning techniques involve training multiple models and aggregating their predictions, which can result in more resilient and versatile results.
Ethical and Bias Concerns
ML and DL models can be biased due to biased training data, model design, and deployment processes. These prejudices can result in unjust and discriminatory results, especially in situations involving people. Ensuring equity and ethical conduct in ML models is a complicated process that involves thorough testing, validation, and monitoring. It is vital to address these concerns in order to avoid harm and guarantee that ML systems are fair and righteous. Dealing with ethical and bias issues necessitates a comprehensive strategy. It is essential to include diversity in training datasets and make a conscious effort to recognize and address biases during data collection and pre-processing. Incorporating algorithms that prioritize fairness and integrating fairness limitations in the model training process can assist in mitigating biased results. It is crucial to constantly monitor and audit models to ensure fairness and reduce bias, as well as to be transparent when reporting the performance of models for various demographic groups. Interacting with ethicists and various stakeholder groups can offer insightful perspectives and direction on the ethical deployment of AI.
Robustness and Adversarial Attacks
ML models, especially DL models, can be easily fooled by adversarial attacks using specially crafted inputs to make inaccurate predictions. These incidents reveal the vulnerability of ML systems and present serious security threats, particularly in vital areas such as autonomous driving and cybersecurity. Research in creating strong models that can resist attacks is ongoing, using methods such as adversarial training and defensive distillation. Nevertheless, developing models that are resilient and precise continues to be a considerable obstacle. Adversarial training can be utilized to boost resilience and protect against adversarial attacks by training models with adversarial examples. Defensive distillation and gradient masking are additional strategies employed to reduce the vulnerability of models to adversarial inputs. Consistently updating models and integrating detection mechanisms to recognize and counter adversarial attacks can enhance resilience. Ongoing studies in durable structures and algorithms that possess natural resistance to adversarial manipulations are showing potential.
Scalability and Real-Time Processing
As data volume rapidly increases, scalability becomes increasingly important for ML and DL systems. Dealing with big datasets and efficiently training models is a challenging task. Moreover, numerous use cases, like real-time analytic and autonomous systems, necessitate quick processing and decision-making skills. It is crucial for the practical deployment of ML models to guarantee they can scale efficiently and function in real-time settings. Approaches such as distributed computing and edge computing are being researched to tackle these issues, yet achieving smooth integration remains complicated. Distributed computing and parallel processing frameworks can help solve problems related to scalability and real-time processing. Methods such as model parallelism and data parallelism enable the distribution of work among various processors. Edge computing, which means handling data close to where it is created (such as IoT devices), can decrease delays and improve the ability to process data in real-time. Enhancing scalability can be improved by implementing efficient algorithms and optimizing code to take advantage of hardware accelerations.
Integration and Deployment
Incorporating ML and DL models into current systems and workflows may present difficulties. Deployment includes not only integrating technically, but also involves thinking about model maintenance, updates, and monitoring. Continuous evaluation and retraining are necessary to maintain the accuracy and reliability of models as new data is introduced. Furthermore, the deployment procedure needs to consider the scalability of the infrastructure, compatibility with other systems, and adherence to regulatory standards. Using containerization technologies like Docker and Kubernetes can aid in the smooth integration and deployment of ML and DL models, allowing for the management and scaling of applications in various environments. CI/CD pipelines automate the deployment process, guaranteeing that models are consistently updated and cared for. Tools for monitoring models are capable of monitoring performance, detecting drifts, and sending alerts to ensure prompt retraining and adjustments. Facilitating smoother integration can be achieved by ensuring that standardized APIs and data formats are used to ensure interoperability with existing systems.
Continuous Learning and Adaptation
ML models must constantly adjust to changing conditions and new data in order to keep up with the dynamic nature of real-world environments. Conventional static models could easily become obsolete, resulting in a decline in performance. Creating systems that can constantly learn and adjust, also known as lifelong learning or online learning, is a difficult yet essential progression for more robust and efficient ML applications. It is essential for these systems to effectively integrate new information while also preserving knowledge obtained in the past. In order to promote ongoing learning and adjustment, incremental learning methods can be utilized, enabling models to be updated with fresh data without needing to start over with training. Online learning algorithms are able to adjust to fresh data instantly, which makes them appropriate for scenarios that are constantly changing. Ensuring that models stay relevant and accurate can be achieved by incorporating feedback loops that use user interactions to inform ongoing model updates. Methods such as transfer learning can be used to utilize knowledge from tasks that have been learned before when facing new tasks that are related.
Cost and Accessibility
The expense of creating and implementing ML and DL solutions may be too high for small and medium-sized businesses (SMEs). The significant expense of computational resources, data acquisition, and skilled staff acts as barriers to entry and restricts the widespread accessibility of these technologies. Decreasing these expenses is important to make ML and DL accessible to more people. This can be achieved through creating better algorithms and providing open-source resources and pre-trained models. Utilizing open-source frameworks and pre-trained models can cut down on costs and improve access to ML and DL technologies, resulting in faster deployment processes. Cloud-based artificial intelligence services provide flexible options that can be utilized on a pay-as-you-use model, allowing smaller businesses to affordably access them. Investing in education and training initiatives for ML and DL can make these technologies accessible to more people. Collaborative platforms and shared resources can also reduce obstacles to getting started.
The Sankey diagram (Fig. 8.2) visually depicts the different obstacles encountered in ML and DL, as well as possible ways to tackle these challenges. The main obstacles consist of problems with data quality, scarcity of training data, worries about data privacy, bias in algorithms, high computational expenses, lack of transparency, compatibility with current systems, difficulties with scaling, and ethical issues. Each of these obstacles is linked to one or more solutions that seek to alleviate the related issues. Data augmentation and transfer learning can help overcome data quality problems and lack of training data by enriching the existing data and boosting model performance. Federated learning addresses data privacy issues by enabling models to be trained on multiple devices without the need to share sensitive data. Regularization techniques help reduce algorithm bias by enhancing both fairness and accuracy. Using cloud computing resources helps to lower computational expenses by offering scalable and cost-efficient computational power. The issue of lacking explainability is tackled with the creation of understandable AI techniques that increase transparency in models' decision-making processes. Utilizing cloud and edge computing can help alleviate integration with existing systems and scalability issues, ensuring efficient processing and seamless integration at the data source. Managing ethical concerns involves implementing ethical guidelines to promote the responsible utilization of AI technologies. Moreover, methods for compressing models aid in decreasing computational burden and enhancing model efficiency.
8.4 Conclusions
ML and DL have been seen remarkable progressions, but it still faces substantial hurdles hindering its full potential. A major hurdle is the requirement for extensive amounts of labeled data to properly train models. Although unsupervised and semi-supervised learning methods are becoming more popular, the need for high-quality labeled data continues to be a limiting factor. Furthermore, the concern regarding data privacy and security has grown more evident, especially as personal and organizational data utilized in training models becomes more sensitive. Approaches such as federated learning and differential privacy are being developed to tackle these issues, although they are still in the early stages of being put into practice. Interpreting and explaining models is crucial, particularly in industries like healthcare and finance, that require transparent and justified decisions based on ML and DL models. Explainable AI (XAI) is advancing, yet finding a middle ground between model complexity and interpretability continues to be a challenge. Moreover, the requirements for computational power and energy consumption for training extensive DL models are significant. Explorations of hardware innovations like improved GPUs and TPUs, along with progress in quantum computing, aim to address these problems, but practical, widely applicable solutions are still in progress.
The ethical challenges posed by AI, such as bias in both data and algorithms, are a major concern. Maintaining fairness and reducing biases necessitates ongoing work in data selection and model development phases. Dealing with these difficulties requires a cross-disciplinary strategy, merging technological progress with strong regulatory structures and moral principles. Advancing in the field requires overcoming these obstacles to ensure the long-term and fair use of ML and DL technologies.
References
Abdar, M., Pourpanah, F., Hussain, S., Rezazadegan, D., Liu, L., Ghavamzadeh, M., ... & Nahavandi, S. (2021). A review of uncertainty quantification in deep learning: Techniques, applications and challenges. Information fusion, 76, 243-297.
Ahmed, S. F., Alam, M. S. B., Hassan, M., Rozbu, M. R., Ishtiak, T., Rafa, N., ... & Gandomi, A. H. (2023). Deep learning modelling techniques: current progress, applications, advantages, and challenges. Artificial Intelligence Review, 56(11), 13521-13617.
Alafif, T., Tehame, A. M., Bajaba, S., Barnawi, A., & Zia, S. (2021). Machine and deep learning towards COVID-19 diagnosis and treatment: survey, challenges, and future directions. International journal of environmental research and public health, 18(3), 1117.
Al-amri, R., Murugesan, R. K., Man, M., Abdulateef, A. F., Al-Sharafi, M. A., & Alkahtani, A. A. (2021). A review of machine learning and deep learning techniques for anomaly detection in IoT data. Applied Sciences, 11(12), 5320.
Alzubaidi, L., Zhang, J., Humaidi, A. J., Al-Dujaili, A., Duan, Y., Al-Shamma, O., ... & Farhan, L. (2021). Review of deep learning: concepts, CNN architectures, challenges, applications, future directions. Journal of big Data, 8, 1-74.
Asharf, J., Moustafa, N., Khurshid, H., Debie, E., Haider, W., & Wahab, A. (2020). A review of intrusion detection systems using machine and deep learning in internet of things: Challenges, solutions and future directions. Electronics, 9(7), 1177.
Bhatt, C., Kumar, I., Vijayakumar, V., Singh, K. U., & Kumar, A. (2021). The state of the art of deep learning models in medical science and their challenges. Multimedia Systems, 27(4), 599-613.
Bhushan, M., Pandit, A., & Garg, A. (2023). Machine learning and deep learning techniques for the analysis of heart disease: a systematic literature review, open challenges and future directions. Artificial Intelligence Review, 56(12), 14035-14086.
Chiche, A., & Yitagesu, B. (2022). Part of speech tagging: a systematic review of deep learning and machine learning approaches. Journal of Big Data, 9(1), 10.
Dar, R. A., Rasool, M., & Assad, A. (2022). Breast cancer detection using deep learning: Datasets, methods, and challenges ahead. Computers in biology and medicine, 149, 106073.
Dinsdale, N. K., Bluemke, E., Sundaresan, V., Jenkinson, M., Smith, S. M., & Namburete, A. I. (2022). Challenges for machine learning in clinical translation of big data imaging studies. Neuron, 110(23), 3866-3881.
Fikri, M. A., Santosa, P. I., & Wibirama, S. (2021). A review on opportunities and challenges of machine learning and deep learning for eye movements classification. In 2021 IEEE International Biomedical Instrumentation and Technology Conference (IBITeC) (pp. 65-70). IEEE.
Fregoso-Aparicio, L., Noguez, J., Montesinos, L., & García-García, J. A. (2021). Machine learning and deep learning predictive models for type 2 diabetes: a systematic review. Diabetology & metabolic syndrome, 13(1), 148.
Goshisht, M. K. (2024). Machine learning and deep learning in synthetic biology: Key architectures, applications, and challenges. ACS omega, 9(9), 9921-9945.
Hong, S., Zhou, Y., Shang, J., Xiao, C., & Sun, J. (2020). Opportunities and challenges of deep learning methods for electrocardiogram data: A systematic review. Computers in biology and medicine, 122, 103801.
Jdey, I., Hcini, G., & Ltifi, H. (2022). Deep learning and machine learning for Malaria detection: Overview, challenges and future directions. arXiv preprint arXiv:2209.13292.
Kaluarachchi, T., Reis, A., & Nanayakkara, S. (2021). A review of recent deep learning approaches in human-centered machine learning. Sensors, 21(7), 2514.
Kaluarachchi, T., Reis, A., & Nanayakkara, S. (2021). A review of recent deep learning approaches in human-centered machine learning. Sensors, 21(7), 2514.
Khademi, Z., Ebrahimi, F., & Kordy, H. M. (2023). A review of critical challenges in MI-BCI: From conventional to deep learning methods. Journal of Neuroscience Methods, 383, 109736.
Kocher, G., & Kumar, G. (2021). Machine learning and deep learning methods for intrusion detection systems: recent developments and challenges. Soft Computing, 25(15), 9731-9763.
L’heureux, A., Grolinger, K., Elyamany, H. F., & Capretz, M. A. (2017). Machine learning with big data: Challenges and approaches. Ieee Access, 5, 7776-7797.
Li, L., Li, X., Jiang, L., Su, X., & Chen, F. (2021). A review on deep learning techniques for cloud detection methodologies and challenges. Signal, Image and Video Processing, 15(7), 1527-1535.
Liu, F., Xue, S., Wu, J., Zhou, C., Hu, W., Paris, C., ... & Yu, P. S. (2020). Deep learning for community detection: progress, challenges and opportunities. arXiv preprint arXiv:2005.08225.
Mishra, R. K., Reddy, G. S., & Pathak, H. (2021). The understanding of deep learning: A comprehensive review. Mathematical Problems in Engineering, 2021(1), 5548884.
Mohammed, A., & Kora, R. (2023). A comprehensive review on ensemble deep learning: Opportunities and challenges. Journal of King Saud University-Computer and Information Sciences, 35(2), 757-774.
Nasir, V., & Sassani, F. (2021). A review on deep learning in machining and tool monitoring: Methods, opportunities, and challenges. The International Journal of Advanced Manufacturing Technology, 115(9), 2683-2709.
Rani, P., Kotwal, S., Manhas, J., Sharma, V., & Sharma, S. (2022). Machine learning and deep learning based computational approaches in automatic microorganisms image recognition: methodologies, challenges, and developments. Archives of Computational Methods in Engineering, 29(3), 1801-1837.
Rezaeianjouybari, B., & Shang, Y. (2020). Deep learning for prognostics and health management: State of the art, challenges, and opportunities. Measurement, 163, 107929.
Sapoval, N., Aghazadeh, A., Nute, M. G., Antunes, D. A., Balaji, A., Baraniuk, R., ... & Treangen, T. J. (2022). Current progress and open challenges for applying deep learning across the biosciences. Nature Communications, 13(1), 1728.
Shafay, M., Ahmad, R. W., Salah, K., Yaqoob, I., Jayaraman, R., & Omar, M. (2023). Blockchain for deep learning: review and open challenges. Cluster Computing, 26(1), 197-221.
Shamshirband, S., Fathi, M., Dehzangi, A., Chronopoulos, A. T., & Alinejad-Rokny, H. (2021). A review on deep learning approaches in healthcare systems: Taxonomies, challenges, and open issues. Journal of Biomedical Informatics, 113, 103627.
Silva, S. H., & Najafirad, P. (2020). Opportunities and challenges in deep learning adversarial robustness: A survey. arXiv preprint arXiv:2007.00753.
Talaei Khoei, T., Ould Slimane, H., & Kaabouch, N. (2023). Deep learning: Systematic review, models, challenges, and research directions. Neural Computing and Applications, 35(31), 23103-23124.
Taye, M. M. (2023). Understanding of machine learning with deep learning: architectures, workflow, applications and future directions. Computers, 12(5), 91.
Thakkar, A., & Lohiya, R. (2021). A review on machine learning and deep learning perspectives of IDS for IoT: recent updates, security issues, and challenges. Archives of Computational Methods in Engineering, 28(4), 3211-3243.
Wani, J. A., Sharma, S., Muzamil, M., Ahmed, S., Sharma, S., & Singh, S. (2022). Machine learning and deep learning based computational techniques in automatic agricultural diseases detection: Methodologies, applications, and challenges. Archives of Computational methods in Engineering, 29(1), 641-677.
Xu, Y., Zhou, Y., Sekula, P., & Ding, L. (2021). Machine learning in construction: From shallow to deep learning. Developments in the built environment, 6, 100045.
Yeole, A. N. (2024). A systematic review on various applications and challenges in deep learning. In AIP Conference Proceedings (Vol. 3112, No. 1). AIP Publishing.
Yuan, Q., Shen, H., Li, T., Li, Z., Li, S., Jiang, Y., ... & Zhang, L. (2020). Deep learning in environmental remote sensing: Achievements and challenges. Remote Sensing of Environment, 241, 111716.
Yuan, S., & Wu, X. (2021). Deep learning for insider threat detection: Review, challenges and opportunities. Computers & Security, 104, 102221.