|
| Ref | Year | Dataset | Results | Contribution | Future work | Limitations |
|
| [14] | 2007 | UCI heart disease dataset contains 13 attributes and 270 instances | Weighted K-NN/87% accuracy | Proposed a cardiac arrhythmia model using K-NN-weighted preprocessing and fuzzy allocation mechanism of artificial immune recognition system | To enhance the model by applying SVM, decision tree, and hybrid techniques | (i) This model identifies only a type of heart disease |
| (ii) The small dataset size is not ideal for stable performance because it results in biased measurements |
| [15] | 2008 | Cleveland heart disease dataset has 909 instances and 15 medical risk features | Naive Bayes/95% accuracy | They developed a risk model using decision tree, neural network, and naive Bayes algorithms | To improve model performance by training on massive data | (i) Small dataset with limited instances |
| (ii) Overfitting problems occur in a small dataset, leading to poor performance with the test data |
| [16] | 2009 | Researchers obtain live datasets from patients with heart disease | Bagging with naive Bayes/84.1% accuracy | Developed risk model using C4.5, bagging with naive Bayes, and bagging with C4.5 | To develop a robust heart disease risk model using python or R | Weka handles small datasets, and whenever a dataset is bigger than a few megabytes, an OutOfMemoryError occurs |
| [17] | 2010 | UCI data repository | The accuracy of the developed model is 82% | (i) Develop fuzzy expert risk model | Build an ensemble model because that will result in a robust and optimal model and increase the model’s efficiency | Derived rules from the cardiovascular disease dataset are complex and large, making the system slow and making wrong decisions |
| (ii) It generated 44 rules compared with the results of other rule bases. |
| [18] | 2011 | Cleveland heart disease dataset has 909 instances and 15 medical risk features | The accuracy of 79.1% with voting and 84.1% without voting is achieved | (i) Develop a classification model using the decision tree | To develop a risk evaluation model using hybrid classification techniques for optimal results | The developed heart disease evaluation models lack generalization ability |
| (ii) Heart disease rules are generated using reduced error pruning |
| [19] | 2012 | Cleveland heart disease dataset | K-NN with the accuracy of 97% | They develop the K-NN model for the early prediction of heart disease | To work on a primary heart disease dataset with considerable volume size | Applying the voting technique did not progress in precision even after estimating different parametric values of k |
| [20] | 2013 | The random dataset of 303 instances is collected | The accuracy of the LAD stenosis is 79.54% | (i) They apply C4.5 and bagging to check the lab and ECG data | To work on a primary heart disease dataset with massive instances using hybrid learning techniques | The model uses invasive risk features, making it difficult for general users and limiting its usage to the medical domain |
| (ii) The Gini index and information gain select the essential features |
| [21] | 2014 | Used Cleveland, Hungarian, and Switzerland datasets | 80% and 42% accuracies on Switzerland and Hungarian heart disease datasets, respectively | (i)Proposed a model by combining rough set theory with the fuzzy set | To develop a risk evaluation model with less computational complexity and high predictive capability | They use medical domain performance measures and do not test the model measures (computational complexity, scalability, robustness, and comprehensibility) |
| (ii) Generate fuzzy base rules using the rough set approach and the fuzzy classifier. |
| [22] | 2015 | Five binary class medical datasets collected from the UCI repository | Optimal results | Initially, the k-means algorithm is used for clustering and then 12 distinct classifiers are used to create the final model using stratified 10-fold cross-validation | To develop a risk model with the best generalization ability and less computational complexity | (i) The developed risk model is complex and takes more computational time |
| (ii) Used Weka tool that handles small datasets, and an OutOfMemoryError occurs on a vast dataset |
| [23] | 2016 | Cleveland heart disease dataset | Accuracy increases by controlling the discrete features using feature selection techniques | Sequential minimal optimization algorithm is applied to develop the risk model using the MATLAB tool | To extend the model by using real-time datasets to get an accurate diagnosis in advance | (i) The model time complexity is high |
| (ii) Overfitting problems occur in a small dataset, leading to poor performance with the test data |
| [24] | 2017 | Used Z-Alizadeh sani heart disease dataset | The accuracy, sensitivity, and specificity are 84%, 85%, and 89% | Proposed a hybrid model that uses the error back propagation algorithm in ANN with MLP structure and sigmoid exponential function | Develop a one-size-fits-all heart disease model to successfully prescribe a treatment plan for the disease | The error generated by the hidden neurons on output nodes degrades the neural network’s logic potential, resulting in wrong prediction and decision-making |
| [31] | 2019 | Review paper | Analyze security models, check trends, and highlight opportunities and challenges for future IoT-based healthcare development | (i) Review latest IoT components, applications, and healthcare market trends | Will address the challenges that prevent the development of IoT and cloud computing in healthcare, such as data security, system development processes, and business models | They did not review IoT privacy and security issues like potential threats, attack types, and security setups |
| (ii) Analyze the influence of cloud computing, big data, and wearables to determine how they help the sustainable development of IoT and cloud computing in the healthcare industry |
| [14] | 2020 | Unknown | MSSO-ANFIS/accuracy = 99.4 and precision = 96.54 | Propose an Internet of Medical Things framework using modified salp swarm optimization and an adaptive neuro-fuzzy inference system for heart disease prediction | Researchers will use different feature selection and optimization techniques to improve the model effectiveness of prediction | The developed heart disease prediction model is complex and expensive because of medical attribute examination and IoT use |
| [32] | 2020 | Live dataset | MDCNN/accuracy = 98.2 | Propose a wearable IoT-enabled framework to evaluate heart disease using a modified deep convolutional neural network (MDCNN) | (i) To increase the model’s performance using other feature selection and optimization techniques | The developed risk model is complex and expensive because of medical attribute examinations and IoT use |
| The MDCNN classifies the received sensor data into normal and abnormal | (ii) To train the model with fully wearable devices available in the market |
| [33] | 2020 | Live dataset | Avg. correlation coefficient is 0.045, encryption time = 1.032S, and decryption time = 1.004 S | Proposed a secure framework that uses the wearable sensor device which monitors blood pressure, body temperature, serum cholesterol, glucose level, etc | To extend this work, such as capturing the data from the wearable sensors and performing real-time analysis | The developed framework is complex because of the use of IoT components |
| [34] | 2021 | Wireless body area networks (WBAN) framework | Execution time, memory, and energy consumption of the developed WBAN are optimal | Propose a three-tier security model for wireless body area networks (WBAN) systems that is suitable for e-health applications | To incorporate the security solutions and concentrate on competitive execution time, memory, and energy consumption | (i) They used lightweight cryptography instead of robust crypto-algorithms |
| (ii) A complete comparison with other methods is difficult due to security services, device types, and security levels. |
| [35] | 2022 | Primary heart disease dataset consisting of 5776 records | Random forest/85% accuracy | Develop an effective, low-cost, reliable risk evaluation model using significant noninvasive risk attributes | (i) To enhance the risk model by adding other noninvasive features | The risk model is developed on a specific population, hence narrowing its application |
| (ii) To investigate deep learning and study the significance of other controlled features on different age and sex groups |
|
