The Internet of Things (commonly referred to as IoT) is a network of connected physical devices – “things” – that are equipped with sensors, software, and other technologies, which enables them to exchange data between each other and the system. At present, the predictive capabilities of machine learning (ML) have been gaining traction in relation to the optimization of IoT performance. When ML algorithms and IoT technologies are well-integrated, it is possible to achieve better control and improved decision-making (Adi et al., 2020). Shanthamallu et al. (2017) argue that all the major machine learning techniques – supervised, unsupervised, and deep learning – can find applicability in the field of IoT. Hussain et al. (2020) justify the use of ML for IoT by pointing out the vast amount of data that IoT gathers. Its volume and variability render conventional data collection, storage, and processing techniques irrelevant and inefficient, which is why it takes more rigorous approaches to address these issues.
The survey conducted by Ciu et al. (2018) reveals more prospects for ML for IoT. The authors enlist anomaly/ intrusion detection, face recognition, voice recognition, malware analysis, and attack detection among the possible applications of ML. One of the most significant things that the paper by Ciu et al. (2018) accomplishes is matching IoT security/ performance problems with ML techniques. For instance, it justifies the use of K-means and K-nearest neighbor clustering for attack detection and mitigation and anomaly/ intrusion detection. Other recent papers, such as the one by Diedrichs et al. (2018), focus on the more specialized use of ML and DL (deep learning) such as the detection of frost events in agricultural IoT. Similarly, Baracaldo et al. (2018) took a specialized approach to adapting ML techniques to IoT and demonstrated its use for detecting poisoning attacks in the environment.
To summarize, based on their purpose, ML analytics techniques for IoT can be put into four distinct categories: (1) descriptive; (2) prescriptive; (3) predictive; and (4) adaptive (Adi et al., 2020). Descriptive analytics helps with data summarization, offering an insight into past events. Historical data is used for making predictions about future events while prescriptive analytics utilize the possibilities of both the first two types to shape an adequate response to events. Lastly, adaptive analytics have the promising potential of adjusting the performance of and calibrating an IoT device based on recent or even real-time data.
Indeed, converging ML techniques with IoT solutions has promising prospects and a wide range of possibilities. However, as pointed out by Pandey (2017), both ML and IoT are emerging fields, though gaining traction exponentially. There is still insufficient research to provide theoretical underpinnings for ML/ IoT solutions, as well as there are industry challenges, to overcome (Pandey, 2017). Li et al. (2018) argue that while ML and IoT are rising trends in the tech world, the supply for adequately sized datasets that would eliminate bias and allow for better fitting models. Though, there are projections that mitigate this barrier: as cited by Adi et al. (2020), by 2022, the world will see 18 billion connected IoT devices. It is readily imaginable that this explosion in use and popularity will account for more ready-to-use data.
Further, the application of ML may fall into one of the traps of big data – its high variability (Mahdavinejad et al., 2018). In real-world IoT environments, data is often unstructured/ semistructured, noisy, and complex. Thus, it is rather challenging to determine which data provides the most meaning and has the highest relevancy. Adi et al. (2020) write that there is an ongoing debate on issues such as sampling from the high-frequency streaming data, filtering, merging of heterogeneous data sources, and its interpretation. Moreover, it is not only about the heterogeneity of data within one setting. Adi et al. (2020) add that different industries have unique needs, and personal smart home data will vary significantly from those gathered at a healthcare facility. Therefore, the question arises as to whether different types of data will pose different storage, processing, and interpreting requirements.
Apart from the aforementioned issues, the techniques themselves also make one question their practicality for IoT solutions. The convergence of ML and IoT is not impeded only by the difficulty of translating ML techniques into practices and their real-time application. For instance, Krishna Sharma and Wang (2018) explain that IoT devices as they are now are often resource-constrained, which means that there exists an optimization issue. No matter how precise, each ML algorithm has an obvious downside: the convergence rate/ learning time. Thus, there is a tradeoff between data transmission and data processing in a limited period of time.
According to Krishna Sharma and Wang (2018), there are design issues with IoT devices that do not allow for easy integration. In particular, Krishna Sharma and Wang (2018) discuss LTE-M and NB-IoT standards that provide only a small bandwidth for IoT, making the issue of the distribution of resources even more acute. To recapitulate, the existing knowledge gaps fall into three categories: (1) the practicality, speed, and efficiency of ML algorithms and learning techniques; (2) the design issues and specifications of IoT devices; and (3) integration issues.
References
Adi, E., Anwar, A., Baig, Z., & Zeadally, S. (2020). Machine learning and data analytics for the IoT. Neural Computing and Applications, 1-29.
Baracaldo, N., Chen, B., Ludwig, H., Safavi, A., & Zhang, R. (2018). Detecting poisoning attacks on machine learning in iot environments. In 2018 IEEE International Congress on Internet of Things (ICIOT) (pp. 57-64). IEEE.
Cui, L., Yang, S., Chen, F., Ming, Z., Lu, N., & Qin, J. (2018). A survey on application of machine learning for the Internet of Things. International Journal of Machine Learning and Cybernetics, 9(8), 1399-1417.
Diedrichs, A. L., Bromberg, F., Dujovne, D., Brun-Laguna, K., & Watteyne, T. (2018). Prediction of frost events using machine learning and IoT sensing devices. IEEE Internet of Things Journal, 5(6), 4589-4597.
Hussain, F., Hussain, R., Hassan, S. A., & Hossain, E. (2020). Machine learning in IoT security: Current solutions and future challenges. IEEE Communications Surveys & Tutorials, 2020, 1-23.
Krishna Sharma, S., & Wang, X. (2018). Towards massive machine type communications in ultra-dense cellular iot networks: Current issues and machine learning-assisted solutions. arXiv, arXiv-1808.
Li, H., Ota, K., & Dong, M. (2018). Learning IoT in edge: Deep learning for the Internet of Things with edge computing. IEEE network, 32(1), 96-101.
Mahdavinejad, M. S., Rezvan, M., Barekatain, M., Adibi, P., Barnaghi, P., & Sheth, A. P. (2018). Machine learning for Internet of Things data analysis: A survey. Digital Communications and Networks, 4(3), 161-175.
Mohammadi, M., Al-Fuqaha, A., Sorour, S., & Guizani, M. (2018). Deep learning for IoT big data and streaming analytics: A survey. IEEE Communications Surveys & Tutorials, 20(4), 2923-2960.
Pandey, P. S. (2017). Machine learning and IoT for prediction and detection of stress. In 2017 17th International Conference on Computational Science and Its Applications (ICCSA) (pp. 1-5). IEEE.
Shanthamallu, U. S., Spanias, A., Tepedelenlioglu, C., & Stanley, M. (2017). A brief survey of machine learning methods and their sensor and IoT applications. In 2017 8th International Conference on Information, Intelligence, Systems & Applications (IISA) (pp. 1-8). IEEE.