An IoT-based blood pressure monitoring system using Support Vector Machine Patience U. Usip1,*,† , Ekrika Kenewenemor2,† and Francis B. Osang2,† 1 Department of Computer Science, University of Uyo, Uyo, Nigeria 2 National Open University of Nigeria, Abuja, Nigeria Abstract The introduction and integration of IoT in daily lives and the general community can be seen in many different astonishing scenarios such as health care. Several medical conditions predominant in patients are often related to high blood pressure. Hence, the need for blood pressure to be closely monitored and the blood pressure made available for IoT devices. Due to the confidentiality and security issues required, machine learning approaches are seen as most suitable. Four machine learning approaches including the support vector machines were used for training the dataset with its discovering patterns and predicting when the patient is in danger. A machine learning-based system for monitoring blood pressure in patients was modeled to provide a reliable blood pressure (systolic and diastolic) prediction and classification to patients. The performance of these models was evaluated using mean square error, mean absolute error, root mean square error, and coefficient of determination (R2). The models were ranked, and the best model selected. This system can accurately predict blood pressure with R2 performance of 0.995551 for SVR, 0.993488 for KNN, 0.993935 for RF, and 0.989557 for CART. This adoption of IoT based blood-pressure monitoring model is encouraged for predicting blood pressure of patients. Keywords Machine learning, medical conditions, Health data,, Data confidentiality and security, 1. Introduction The introduction of IoT medical devices has tremendously transformed the healthcare continuum on an unprecedented scale with many noticeable benefits. The IoT is one of many technologies that has gradually stepped into the health care domain and has provided renewed and improved delivery of healthcare services. With these IoT healthcare monitoring devices, patients may no longer require making regular visits to the hospital as the devices are designed to monitor vital signs of patients anywhere, they are in the universe and data collected are uploaded to a cloud repository, where a remote doctor or nurse can examine for further analysis. The IoT’s real-time monitoring empowers doctors to make actionable insight on data on a continuous basis. This facilitates early detection of disease before it aggravates. These devices are easy AISD-2024: Second International Workshop on Artificial Intelligence: Empowering Sustainable Development, October 2, 2024, co-located with the Second International Conference on Artificial Intelligence: Towards Sustainable Intelligence (AI4S-2024), Virtual Event, Lucknow, India. Year: 2024. * Corresponding author. † These authors contributed equally. $ patienceusip@uniuyo.edu.ng (P. U. Usip); ekrikakenny@gmail.com (E. Kenewenemor); fosang@noun.edu.ng (F. B. Osang)  0000-0002-6516-5194 (P. U. Usip); 0000-0002-2111-5785 (E. Kenewenemor); 0009-0000-7301-4866 (F. B. Osang) © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). CEUR ceur-ws.org Workshop ISSN 1613-0073 Proceedings to use and are available in many different forms: smart watches, eyeglasses, belts etc. they are easy to use and are capable of tracking on real-time data generated from the body such as temperature, sleep, EEG, blood sugar, heart rate, pulse rate, BP levels although the list is not exhaustive. The incidence of cardiovascular disease is commonplace in our local community. For this reason, this has perhaps led to early deaths in recent times. Therefore, the use of these wearable medical devices can provide a medical support system for patients managing heart disease. More so, the use of these collaborative smart medical devices and Machine learning techniques can provide health status forecasts and give doctors insight to prevent heart attack, stroke or worse case death in patients.[1] 2. Recent Related Works Khan et al. [2] proposed an IoT algorithm for forecasting whether the patient under close observation is in undergoing stress or not by monitoring his/her heartbeats. The system is designed to detect the pulse signals and waveform using a specifically dedicated Wi-Fi equipment board that sends data to a server repository. Next, the raw data garnered at different intervals are assembled and stress prediction is evaluated by applying ML techniques as seen in [3] and others such as Support Vector Machine (SVM) and Logistic regression is applied. Results from the simulation tests showed that the precision of the proposed framework can reach up to 68Verma P and Sook [4] proposed an approach for active monitoring of various diseases. It predicts the level of severity of these diseases from normal to severe in a sample data using students. The concept of computational science on the data was garnered from the student using low power sensors. The data is collected and stored in the repository for analysis to predict the severity of the disease. Additionally, the approach used various ML algorithms for classification to forecast the occurrence of such diseases. Although other intelligent approaches have been adopted in Health other than machine learning [5]. The result was evaluated using various metrics such as T-measuring, specifications, and sensitivity. Consequently, the simulation result shows that in terms of correction and precision, the purported methodology surpassed the orthodox approach. The authors Kumer and Gandhi [6] developed a framework of three-layered architecture to deposit or cache enormous sensory data for fast prediction of heart diseases. In the proposed framework, the foremost layer is the data collection layer. The second layer is the repository layer where data is stored in the cloud. The final layer is the third layer, a forecast model is designed to predict the likelihood of heart diseases. Also, at this layer, analysis is performed to spot potential symptoms of heart disease before they occur. On any wearable device, the data collected is transmitted to an edge computing network as the wearable device is not able to hold an enormous amount of data. Geloyo et al. [7] presented a novel framework of a machine learning-based model for disease classification used in the health monitoring of patients. The simulation is done using this methodology produced accurate results in detecting diseases. Verma et al. [8] proposed a smart student m-healthcare monitoring framework based on cloud-centric IoT is suggested. This framework calculates the severity of a student’s ailment by temporally mining the health metrics obtained from medical and other IoT devices to anticipate the probable disease and its level. An architectural model for a smart student health care system has been built to properly evaluate student healthcare data. In this case study, we used a health dataset of 182 suspected students to construct instances of waterborne diseases. Using the k-cross-validation technique, this data is further evaluated to validate the model. Various classification algorithms are used to apply a pattern-based diagnosis scheme, and the results are then computed based on accuracy, sensitivity, specificity, and reaction time. In terms of the above-mentioned parameters, experimental findings reveal that decision tree and k-nearest neighbor algorithms outperform other classifiers. Furthermore, the suggested technique aids decision-making by providing time-sensitive information to the caretaker or doctor at a given moment. Finally, the suggested system’s temporal granule pattern-based presentation yields good diagnosis findings. Based on Internet-of-Things (IoT) technology, Zhe Yang1 et al. [9] presented a novel approach for ECG monitoring. ECG data is collected via a wearable monitoring node and wirelessly transferred to the IoT cloud. The Hypertext Transfer Protocol (HTTP) and MQ Telemetry Transport (MQTT) protocols are used in the IoT cloud to offer users visible and fast ECG data. The cross-platform difficulty has been significantly reduced because nearly all smart terminals with a web browser can easily obtain ECG data. Experiments on healthy volunteers are carried out to ensure that the entire system is reliable. The suggested system is reliable in collecting and presenting real-time ECG data, which can help in the primary diagnosis of some cardiac conditions, according to experimental results. 3. Methodology The research methodology focuses on the use of blood pressure datasets obtained from an online repository of patient’s vital signs, the approach will employ SVR and CART for data classifications and relations. Afterwards, train a model leveraging ML algorithm and the dataset obtained as describe above. The overall result of this methodology is to design a system that can correctly predict spikes in blood pressure levels and inform the relevant healthcare providers or givers on real time. Lastly, the system will be evaluated to measure correctness and accuracy of the system 3.1. Proposed System Framework This work proposes a machine learning-based system for monitoring blood pressure in patients. The work employs a mix of machine learning models namely support vector regression (SVR), random forest (RF), k-nearest neighbor (KNN), and classification and regression tree (CART) to predict patient blood pressure level. Machine learning algorithms learn from datasets to predict patient’s blood pressure employing the blood pressure dataset obtained from kaggle.com. The blood pressure dataset is partitioned into training set and testing set. The training set is made up of 70To select the best machine learning model, model evaluation is carried out using standard regression-based machine learning model evaluation techniques such as MSE (Mean Square Error), MAE (Mean Absolute Error), and RMSE (Root Mean Square Error). The proposed framework of machine learning-based system for monitoring blood pressure in patients is presented in Figure 1. The system framework depicts the structure, components and the relationship among the various components making up this research work. Figure 1: System Framework 4. Model Performance Result and Discussions The ability of each machine learning model to accurately predict the blood pressure given a new dataset is evaluated using Mean Square Error, Root Mean Square Error, Mean Absolute Error, and Coefficient of Determinant (R2). The performance of these models is presented in Table 2. From Table 2, it is observed that SVR has the best performance/accuracy in predicting the blood pressure by RF, KNN then CART. The performances of these models are presented graphically in Figure 2. From Figure 2, it is observed that SVR outperforms all other models in terms of MSE, RMSE, MAE, and R2. Hence Support Vector Regression is recommended for effective prediction of blood pressure in patients. Model Selection The best model is a model with the highest coefficient of determination. The graph in Figure 3, ranks our machine learning model in order of its performance. From Figure 3, SVR is the best model with R2 of 0.995551. The interpretation of the SVR model result is presented in Table 3 It shows the predicted severity of blood pressure. 5. Conclusion A machine learning-based system for monitoring blood pressure in patients was modeled to provide a reliable blood pressure (systolic and diastolic) prediction and classification to patients. After an in-depth review of the study area, machine learning models – support vector machine, random forest, k-nearest neighbor, and classification and regression tree models where developed. These models where trained and tested using the blood pressure dataset. The performance of these models was evaluated using mean square error, mean absolute error, root Table 1 Model Predictions: SVR Temp rr % Hr % diastolic % systolic % SVR diastolic SVR systolic 29.98063 2 30 57 16 61.32608 11.09031 39.97486 2 59 16 115 15.51658 119.2956 29.51347 33 103 63 16 59.47146 11.0465 20.32661 24 73 7 3 10.16969 7.594763 28.08248 43 107 63 168 65.39091 166.0555 29.91006 21 34 59 75 61.40598 74.60439 23.43972 16 27 106 148 105.4465 147.3131 20.519 38 146 144 178 140.5831 176.1364 32.98457 11 133 133 144 128.8536 146.0056 28.74359 23 98 6 87 6.253409 84.05716 34.378 50 162 118 43 118.6306 43.42842 25.48815 1 64 26 92 26.46416 94.8899 38.93474 3 12 17 151 16.47869 152.0219 30.79187 7 148 106 101 108.2135 96.75982 34.46068 52 138 129 106 124.5589 102.0308 20.92426 21 88 139 49 140.9551 44.47636 25.31142 11 134 90 194 89.41622 194.3509 37.71674 42 185 118 65 121.199 63.0154 31.55399 53 35 91 32 91.75912 31.95917 32.42716 43 162 97 88 92.60312 83.23546 36.4954 26 3 80 81 78.78053 81.11791 37.86951 17 166 136 122 139.7972 124.0788 34.48373 0 26 118 67 118.0558 67.86683 22.25193 9 192 138 132 138.4186 127.6537 mean square error, and coefficient of determination (R2). The models were ranked, and the best model selected. This system can accurately predict blood pressure with R2 performance of 0.995551 for SVR, 0.993488 for KNN, 0.993935 for RF, and 0.989557 for CART. References [1] D. S. Rajput, R. Gour, An iot framework for healthcare monitoring systems, International Journal of Computer Science and Information Security 14 (2016). [2] F. Khan, M. A. Jan, A. ur Rehman, S. Mastorakis, M. Alazab, P. Watters, A secured and intelli- gent communication scheme for iiot-enabled pervasive edge computing, IEEE Transactions on Industrial Informatics 17 (2020) 5128–5137. [3] P. Naveen, M. Maragatharajan, R. Thangaraj, Contextual reinforcement learning for en- hanced machine translation using transformers., in: SWIoT+ MSW@ KGSWC, 2023, pp. 39–50. [4] P. Verma, S. K. Sood, A comprehensive framework for student stress monitoring in fog-cloud Table 2 Model Performance on blood pressure dataset MSE RMSE MAE R2 SVR 8.656671 2.942222 2.546986 0.995551 RF 11.80007 3.435124 2.972931 0.993935 CART 20.31929 4.507692 3.917981 0.989557 KNN 12.67081 3.559607 3.064961 0.993488 Figure 2: Model performance graphs iot environment: m-health perspective, Medical & biological engineering & computing 57 (2019) 231–244. [5] P. U. Usip, M. E. Ekpenyong, F. F. Ijebu, K. J. Usang, Integrated context-aware ontology for mnch decision support, in: Semantic Models in IoT and eHealth Applications, Elsevier, 2022, pp. 227–243. [6] P. M. Kumar, U. D. Gandhi, A novel three-tier internet of things architecture with machine learning algorithm for early detection of heart diseases, Computers & Electrical Engineering 65 (2018) 222–235. [7] Y. E. Gelogo, J.-W. Oh, J. W. Park, H.-K. Kim, Internet of things (iot) driven u-healthcare sys- tem architecture, in: 2015 8th International Conference on Bio-Science and Bio-Technology (BSBT), IEEE, 2015, pp. 24–26. [8] P. Verma, S. K. Sood, S. Kalra, Smart computing based student performance evaluation Figure 3: Model Ranking framework for engineering education, Computer Applications in Engineering Education 25 (2017) 977–991. [9] Z. Yang, Q. Zhou, L. Lei, K. Zheng, W. Xiang, An iot-cloud based wearable ecg monitoring system for smart healthcare, Journal of medical systems 40 (2016) 1–11. Table 3 Interpretation of the SVR model result Temp rr Hr diastolic systolic SVR (diastolic) SVR (systolic) Severity 29.98063 2 30 57 16 61.32608 11.09031 Low 39.97486 2 59 16 115 15.51658 119.2956 Ideal 29.51347 33 103 63 16 59.47146 11.0465 Low 20.32661 24 73 7 3 10.16969 7.594763 Low 8.08248 43 107 63 168 65.39091 166.0555 High 29.91006 21 734 59 75 61.40598 74.60439 Low 23.43972 16 27 106 148 105.4465 147.3131 High 20.519 38 146 144 178 140.5831 176.1364 High 32.98457 11 133 133 144 128.8536 146.0056 High 28.74359 23 98 6 87 6.253409 84.05716 Low 34.378 50 162 118 43 118.6306 43.42842 High 25.48815 1 64 26 92 26.46416 94.8899 Ideal 38.93474 3 12 17 151 16.47869 152.0219 High 30.79187 7 148 106 101 108.2135 96.75982 High 34.46068 52 138 129 106 124.5589 102.0308 High 20.92426 21 88 139 49 140.9551 44.47636 High 25.31142 11 134 90 194 89.41622 194.3509 High 37.71674 42 185 118 65 121.199 63.0154 High 31.55399 53 35 91 32 91.75912 31.95917 High 32.42716 43 162 97 88 92.60312 83.23546 High 36.4954 26 3 80 81 78.78053 81.11791 Ideal 37.86951 17 166 136 122 139.7972 124.0788 High 34.48373 0 26 118 67 118.0558 67.86683 High 22.25193 9 192 138 132 138.4186 127.6537 High