Access to this full-text is provided by Tech Science Press.
Content available from Computers, Materials & Continua
This content is subject to copyright. Terms and conditions apply.
This work is licensed under a Creative Commons Attribution 4.0 International License,
which permits unrestricted use, distribution, and reproduction in any medium, provided
the original work is properly cited.
ech
T
PressScience
Computers, Materials & Continua
DOI: 10.32604/cmc.2023.033603
Article
Feature-Limited Prediction on the UCI Heart Disease Dataset
Khadijah Mohammad Alfadli and Alaa Omran Almagrabi*
Department of Information Systems, Faculty of Computing and Information Technology, King Abdulaziz University,
Jeddah, 21589, Saudi Arabia
*Corresponding Author: Alaa Omran Almagrabi. Email: aalmagrabi3@kau.edu.sa
Received: 22 June 2022; Accepted: 11 October 2022
Abstract: Heart diseases are the undisputed leading causes of death globally.
Unfortunately, the conventional approach of relying solely on the patient’s
medical history is not enough to reliably diagnose heart issues. Several poten-
tially indicative factors exist, such as abnormal pulse rate, high blood pres-
sure, diabetes, high cholesterol, etc. Manually analyzing these health signals’
interactions is challenging and requires years of medical training and experi-
ence. Therefore, this work aims to harness machine learning techniques that
have proved helpful for data-driven applications in the rise of the artificial
intelligence era. More specifically, this paper builds a hybrid model as a tool
for data mining algorithms like feature selection. The goal is to determine
the most critical factors that play a role in discriminating patients with heart
illnesses from healthy individuals. The contribution in this field is to provide
the patients with accurate and timely tentative results to help prevent further
complications and heart attacks using minimum information. The developed
model achieves 84.24% accuracy, 89.22% Recall, and 83.49% Precision using
only a subset of the features.
Keywords: Machine learning; feature selection; heart disease
1Introduction
According to the World Health Organization (WHO), cardiovascular diseases (CVDs), commonly
known as heart diseases, are the leading causes of death globally. In 2016, the total death count
reached about 58 million people, 31% of whom died due to CVDs. Most of these deaths, around 85%,
were from heart attacks and strokes [1]. WHO has put a worldwide action plan spanning 2013 to
2020 in response to CVDs and cancer, diabetes, and chronic respiratory diseases, collectively known
as Noncommunicable diseases (NCDs). The goal is to attain a 25% relative reduction in premature
death from NCDs by 2025. These efforts are necessary steps toward fighting this on a global scale.
However, humanity needs awareness on a more individual level. For example, the American Health
Association has reported several behavioral risk factors that can be regulated to prevent CVDs, such
as smoking cigarettes, eating unhealthy food, and not exercising regularly. High cardiovascular risk
patients suffering from hypertension, diabetes, and/or hyperlipidemia should be closely monitored as
early detection of CVDs can prevent premature deaths [2]. Many of these risk factors can be easily
5872 CMC, 2023, vol.74, no.3
measured using accessible tools that might be part of any modern household. Moreover, with the
advancement of technology, there are even now smartwatches and wearables equipped with health-
tracking sensors. Every factor alone might not be a good indicator of heart disease, but their interaction
can provide a clearer signal to the health counselor or the doctor [3]. Developing systems that can assist
human professionals in monitoring high-risk patients is a good strategy for performing widespread
testing for CVDs and devising proactive measures [4]. To enable such application without losing utility,
it is imperative to use as less information about the patient as possible.
With the rise of the Artificial Intelligence (AI) era, many data-driven problems have become
possible to solve with expert accuracy. Most of the recent success can be attributed to the advances
in Machine Learning (ML), a subfield of AI that relies heavily on abundant data. To that end, using
datasets that contain patients’ information with and without CVDs, such as the UCI Heart Disease
Dataset [5], is essential to applying ML algorithms. However, analyzing these datasets requires cleaning
and preprocessing. This work proposes different approaches to classify early whether a patient has
heart disease using classical and modern ML methods with the help of some Data Mining (DM)
techniques. Finally, it will perform ablation studies to determine the most distinctive features of CVDs.
2Related Work
To reliably diagnose heart diseases in a patient, a doctor needs to ask some questions and run a few
tests. The goal is to identify important attributes as the basis for the final diagnosis. Examples include
the patient’s age, sex, type of chest pain, and resting blood pressure. In the ML community, these
attributes are referred to as features. One can formulate the problem as an ML problem (precisely,
a classification problem); given the input features (i.e., patient information), the goal is to predict
whether the patient has cardiovascular disease (CVD) or not. The proposed system attempts to solve
this problem to prevent further complications that might lead to heart failures like heart attacks and
strokes [4].
2.1 Algorithms
From the DM field [6], it is known that some features are more important than others for
classification. However, sometimes the combination of two weak features can be more critical than a
stronger feature. All of this led to the study of feature selection methods [7]. Examples of such methods
include the Relief method, Minimal-Redundancy Maximal-Relevance Algorithm, Least Absolute
Shrinkage, and the Selection Operator, all of which were studied for CVDs in [8]. This research will
leverage feature selection to its advantage for two main reasons. The first reason is to improve the
predictive power of the proposed classifier. The second reason is to train multiple models that rely on
less information which helps when specific values are hard to attain (e.g., blood pressure is unknown).
ML classifiers can be divided into classical and modern [9]. Heart disease prediction systems
were developed using both methods. Examples of classical methods include K-Nearest Neighbor
(KNN), Support Vector Machine (SVM) [10], and Naive Bayes (NB) [11], all of which were studied
in [12]. Other classical approaches include Logistic Regression (LR) [13], Ridge Classifier (RC) [14],
Linear Discriminant Analysis (LDA) [15], Gaussian Process (GP) [16], Decision Tree (DT) [17], and
Random Forest (RF) [18]. Modern ML methods focus on Deep Learning (DL), the study of deep
Artificial Neural Networks (ANN). Examples of ANNs include Multi-Layer Perceptron (MLP) [19]
and Recurrent Neural Network (RNN) [20]. This paper will compare a few classical and modern ML
methods and build a hybrid model combining multiple models, also known as the ensemble model,
as in [21]. Ensembles are better since two minds are always better than one (e.g., the wisdom of the
CMC, 2023, vol.74, no.3 5873
crowd). The biggest hurdle to this work is the availability of data. Since health records are considered
private information, coming across useful data for research is not as easy as in other fields. Up to
our knowledge, the only publicly available dataset for CVD was collected three decades ago [5]. Other
datasets exist, but they require signing NDAs because of their sensitive nature. Hence, most cited work
use only this dataset [22].
2.2 Contributions
The main contributions can be summarized as follows: (1) Providing exploratory data analysis on
the UCI Heart Disease Dataset to study its features. (2) Following proper ML workflow to train on the
entire dataset without removing patients with missing values. (3) Determining the most discriminative
features of CVDs using feature selection on an ensemble model. (4) Performing a comparative study
of multiple ML models and releasing a competitive model using only a few selected features. (5) Open-
sourcing reproducible code for all the experiments in the supplementary material. Contemporary arts
exist, such as [19]and[23]. Nevertheless, they do not train on the entire dataset and do not perform
feature selection.
3UCI Heart Disease Dataset
This dataset was collected in 1988 from four cities: Cleveland, Hungary, Switzerland, and Long
Beach. It has 920 cases of people with and without CVDs with 76 attributes each. However, only 13
are used in practice, as seen in Fig. 1. The top five plots illustrate the histograms for the numerical
features in the dataset. The count of patients missing the value for a particular feature is presented in
the legend. The bottom plots show the categorical features in pie charts (missing values are labeled as
“Unknown”).
After the Extract-Transform-Load (ETL) step comes performing Exploratory Data Analysis
(EDA). The first thing to note here is that two-thirds of the cases have missing values. Removing
them as commonly practiced is unadvisable since the dataset is already too small. In addition, the
data shows five different CVDs severity levels ranging from healthy to Severe. These class labels are
imbalanced, but it is possible to balance them out by changing the problem into binary classifications
(two class labels: healthy and unhealthy). From this point onward, this assumption will be held to
simplify the analysis. Lastly, it is essential to mention that about 80% of the patients are males. This
is unlikely a truly representative sample of the real world, which might indicate a bias in the dataset.
It is paramount to keep this in mind as it might have a detrimental effect on the predictive power and
reliability of the trained models.
Nevertheless, one needs to see how the data is distributed given the class label to get a deeper
insight into interpreting these features. Fig. 2 plots the numerical features against each other in pairs
while color coding the points by whether the patients suffer from CVDs or not (missing values
are ignored). The plots on the diagonal are simply the histograms of the features since the scatter
plot of any feature will result in a degenerate line. It can be observed that the most discriminative
features are “Max Heart Rate” and “ST Depression Peak”. In addition, there is no strong correlation
between the features, which means that they encode different information and are not replaceable (no
multicollinearity).
5874 CMC, 2023, vol.74, no.3
Figure 1: UCI heart disease dataset features’ distributions
CMC, 2023, vol.74, no.3 5875
Figure 2: Numerical features’ correlations for patients with and without CVDs
The same analysis can be applied to the categorical features, as shown in Fig. 3.Fromthese
plots, it can be observed that the ratio of healthy to sick people is almost five times more in males
than females. This could be true globally, but one cannot be confident of this since the data is not
statistically significant. Furthermore, most patients with heart disease appear to have no chest pain,
“Asymptomatic”, which shows the importance of this research. Cases like this can easily go unnoticed
and undiagnosed.
Figure 3: (Continued)
CMC, 2023, vol.74, no.3 5877
4.1 Data Preprocessing
4.1.1 Numerical Features
Since every feature has different ranges, like age and heart rate, this work will apply normalization
to them. Normalization can significantly impact the trained model as it avoids preempting it to think
that heart rate is more important than the person’s age. If such a relationship exists, the model should
learn it on its own. To that end, the experiments will normalize the features to be zero-centered with
unit variance. This is done by taking the mean μand dividing by the standard deviation σ;x←x−μ
σ.
4.1.2 Categorical Features
Most ML models work strictly with numerical features. It is possible to convert categorical features
into numerical features. The trick is to use one-hot encoding, an all-zeros vector with a single element
being one corresponding to the index of the category.
4.1.3 Missing Values
This work replaces any missing value with the mean if it was numerical or a new class label
“Unknown” if it was categorical, and the target classes are balanced by repeating randomly selected
cases.
4.1.4 Data Splits
Training a complex model on simple data can result in overfitting. Informally, it is when the model
can memorize the dataset entirely without learning how to classify it correctly. It is the model’s inability
to detect the underlying patterns in the data. Whereas training a simple model on complex data might
result in underfitting (learning trivial rules). For example, a model classifies patients based on age
only (sick if old and healthy otherwise). To avoid both problematic outcomes, the data is split into
two chunks. The first split will be used to train the model, and the second to test it. Both splits should
be representative enough of the entire dataset (the same ratios of healthy to sick cases; stratified). A
trained model is overfitting if its performance in training surpasses the testing and underfitting if it
could not improve over a fixed classifier; it always predicts the same thing (healthy or sick) regardless
of the input.
4.2 Training
4.2.1 Hyperparameter Tuning
Each ML model has a few configurable hyperparameters, a set of properties that changes its
training behavior and final performance. Usually, they depend on each other (e.g., a particular
hyperparameter setting has a different meaning and effect if the value of another hyperparameter is
changed). So, to achieve the best results for a model, it needs to be trained under all combinations of
possible assigned values for its hyperparameters if feasible. This is what is known as hyperparameter
tuning through grid-search. Tabl e 1 lists the models and their grid-search values. However, it is possible
to accidentally face overfitting on the test set during hyperparameter tuning (hold-out set leakage).
Therefore, it is a widespread practice to tune on a small chunk of the training set, usually called the
validation set.
5878 CMC, 2023, vol.74, no.3
Table 1: ML models and their grid-search values for hyperparameter tuning
Model Hyperparameter Grid-Search Values
KNN n_neighbors 1, 2, 3, 4
weights ‘uniform’, ‘distance’
p1,2,3
LR penalty ‘l1’, ‘l2’
C 100, 10, 1, 0.1, 0.01, 0.001
SVM kernel ‘linear’, ‘rbf’
NB alpha 1, 0.1, 0.01, 0.001, 0.0001, 0.00001
DT max_features ‘auto’, ‘sqrt’, ‘log2’, 5, 10, 30
max_depth 2, 8, 16, 32, 64, 128
min_samples_split 1, 2, 4, 8, 16, 24
min_samples_leaf 1, 2, 5, 10, 15, 30
RF n_estimators 10, 50, 100, 200, 500
max_features ‘auto’, ‘sqrt’, ‘log2’, 5, 10, 30
max_depth 2, 8, 16, 32, 64, 128
min_samples_split 1, 2, 4, 8, 16, 24
min_samples_leaf 1, 2, 5, 10, 15, 30
RC alpha 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
GP max_iter_predict 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000
LDA solver ‘lsqr’, ‘eigen’
shrinkage ‘empirical’, ‘auto’, 0.0001, 0.001, 0.01, 0.1, 0.5, 1
MLP1num_layers 1, 3, 5
max_units 30, 50, 100
batchnorm False, True
dropout 0, 0.5
RNN2hidden_size 30, 50, 100
num_layers 1, 2
bidirectional False, True
dropout 0, 0.3
model ‘GRU’, ‘LSTM’, ‘RNN’
1DL models (MLP and RNN) are trained using the default Adam optimizer.
2RNN treats the features as a fixed sequence processing them one-by-one followed by a linear classifier head.
4.2.2 K-Fold Cross-Validation
Sometimes it is not clear what constitutes a representative training-validation split. One technique
is randomly splitting the training data into K chunks of equal sizes. Then, pick a single chunk for
validation, and the rest K −1 chunks are combined as the training set. After that, repeat the same
thing for all K chunks (every time, the validation set is a different chunk). The final score is averaged
over all the K trials. This score will reflect the true strength of the model and training procedure better
CMC, 2023, vol.74, no.3 5879
than an arbitrary splitting. This method is what is known as k-fold cross-validation in the ML field.
It is necessary to use this score for hyperparameter tuning and only report the performance on the
untouched testing set.
4.3 Model Ensemble
After building and training multiple ML models, one can compare their classifying power and see
which one performs better. However, some classifiers may better classify certain types of patients than
others. Therefore, choosing which classifier works best in every situation is possible. This method is
an example of a model ensemble, which combines the prediction of multiple classifiers to get at least a
better accuracy than all constituent classifiers. This paper will combine the top five scoring classifiers
in the experiments to build a powerful ensemble model.
4.4 Feature Selection
Some features are more relevant than others. For instance, the number of languages a patient
can speak has no relation to whether they have heart disease or not. In addition, ML models can be
susceptible to noise. Moreover, some features might not be easily attainable or measured. For instance,
not every patient is willing to spend money on an MRI scan. A simple feature selection technique
can be used; select the most correlated features to the output. One well-known scoring function
for the features is the χ2Test (pronounced as chi-square). It can measure how likely two random
variables are to be independent. It is a univariate feature selection method and does not consider the
cumulative effect of all the features. For example, two weak features together can be better than a single
strong feature. It also ignores correlated features. Therefore, this work will also apply feature selection
with multiple variables using Importance Permutation [24]. The earlier method tests the dependence
between every feature and the target, while the latter uses a trained model to evaluate the effect of
introducing randomness to every feature.
5Experiments
5.1 Setup
The classical ML models will be trained using the PyCaret package [25], and the DL models
will be implemented using the PyTorch package [26]. Moreover, this work has developed an interface
between the two packages to facilitate the training and the analysis using the convenient features of
PyCaret. It is worth noting that the code is modular and can be easily adapted to work with any
other PyTorch classifier. The experiments will be done under Python 3.9 in Google Colaboratory [27].
Under the hood, NumPy [28], Pandas [29], Scikit-Learn [30], Matplotlib [31], and Seaborn [32]are
used as supporting libraries. The dataset will be divided into 80% as the training set and 20% as the
testing set. Finally, seven evaluation metrics will be reported per model on the testing set3. These exact
experiments will be repeated twice, once on the complete feature set and again on the selected subset
of features.
5.2 Results on the Complete Features
Table 2 only represents the top five machine learning models, followed by the top two deep learning
models and the model ensemble. It can be noticed that the models with the best overall performance
on the complete feature set are the more complex models (the best value for each metric is in boldface
3The description of the models and evaluation metrics were omitted since they are not part of the contribution.
5880 CMC, 2023, vol.74, no.3
font). Here, the ensemble model is the most accurate, achieving 83.15% accuracy on the testing set. It is
essential to mention that this number cannot be compared directly with other sources as this work uses
the entire dataset here, including the patients with missing values. Finally, the MLP model performs
the best in the AUC metric over all the other models.
Table 2: Models’ test comparisons trained using the complete set of features
Model Accuracy AUC Recall Precision F1 Kappa MCC
SVM 83.15 89.47 87.25 83.18 85.17 65.70 65.80
GP 80.43 86.24 85.29 80.56 82.86 60.12 60.25
LR 80.43 88.85 83.33 81.73 82.52 60.31 60.32
RF 79.35 88.83 83.33 80.19 81.73 58.00 58.06
RC 78.80 00.00 79.41 81.82 80.60 57.26 57.29
MLP 82.61 90.58 85.29 83.65 84.47 64.72 64.73
RNN 80.98 88.30 82.35 83.17 82.76 61.55 61.55
Ensemble 83.15 89.97 86.00 83.97 84.95 65.83 65.90
5.3 Feature Selection
Fig. 5 compares both feature selection methods, χ2Test and Importance Permutation, where
the latter was computed using the best model from the previous experiments. Despite their distinct
ordering, one can see that the top seven features are shared among the two rankings. The same goes
for the bottom features, where both methods agree that they are mostly irrelevant to the task at
hand. Following best intuition, the top seven features will be selected for the ablation study of feature
selection [33].
Figure 5: Feature importance rankings using two different methods
5.4 Results on the Selected Features
Table 3 shows that the best model for the selected features is the GP model. It is even better than
all other models, including the deep learning and the ensemble model. More interestingly, the achieved
accuracy is better than using the full features. Two things can explain this; the feature selection done
CMC, 2023, vol.74, no.3 5881
on the best model on the complete feature set did its job ideally, and the GP model is more suitable
when using fewer data under more uncertainty. The final observation is that the RNN model was not
better in both experiments. However, its striking consistency despite using vastly different features
demonstrates its robustness. This can mean that it has played a vital role in the model ensemble in
Table 2.
Table 3: Models’ test comparisons trained using the selected set of features
Model Accuracy AUC Recall Precision F1 Kappa MCC
GP 84.24 87.45 89.22 83.49 86.26 67.83 68.04
LR 80.43 88.49 82.35 82.35 82.35 60.40 60.40
SVM 80.43 88.70 87.25 79.46 83.18 59.92 60.30
RC 79.89 00.00 80.39 82.83 81.59 59.45 59.48
LDA 79.89 88.40 81.37 82.18 81.77 59.35 59.35
MLP 81.52 88.73 85.29 82.08 83.65 62.42 62.48
RNN 80.98 88.07 89.22 79.13 83.87 60.89 61.55
Ensemble 81.12 89.00 81.81 83.85 82.75 61.89 62.03
5.5 Limitations and Future Work
The most significant limitation of this work is the non-sufficient data to draw statistically
significant conclusions. This research can benefit greatly from more rich and diverse datasets publicly
available for general use. To build on this work, one can collect more data and introduce new features
that can improve the discriminative power of these classifiers. For example, the developed system can
be used to collect anonymized data for similar future applications. Moreover, it might be interesting to
predict the hardest available features using other more attainable information, extending the usability
to a broader audience.
6Conclusion
This research tackled the prediction problem of the UCI heart disease dataset in a feature-limited
setting. It followed a proper data science workflow from data analysis and preprocessing to model
building, training, and evaluation. In particular, this work trained multiple classical machine learning
and deep learning models including a hybrid model of all the top performing models. Each model was
tested under different hyperparameter configurations using a validation data split. Then, this paper
applied feature selection and repeated the same process to get a model that uses only a subset of the
features with competitive performance. This makes it easier for patients with limited access to benefit
from the system while achieving an 84.24% accuracy, 89.22% Recall, and 83.49% Precision. As a result,
this effort satisfied two critical goals of machine learning: interpretation and prediction.
Funding Statement: The authors received no specific funding for this study.
Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the
present study.
5882 CMC, 2023, vol.74, no.3
References
[1] World Health Organization, “The top 10 causes of death,” 2020. [Online]. Available: https://www.who.int/
news-room/fact-sheets/detail/the-top-10-causes-of-death.
[2] E. J. Benjamin, P. Muntner, A. Alonso, M. S. Bittencourt, C. W. Callaway et al., “Heart disease and stroke
statistics—2019 update: A report from the American heart association,” Circulation, vol. 139, no. 10, pp.
56–528, 2019.
[3] S. Bashir, Z. S. Khan, F. H. Khan, A. Anjum and K. Bashir, “Improving heart disease prediction using
feature selection approaches,” in Proc. of IBCAST, Islamabad, Pakistan, pp. 619–623, 2019.
[4] A. H. Chen, S. Y. Huang, P. S. Hong, C. H. Cheng and E. J. Lin, “HDPS: Heart disease prediction system,”
in 2011 Computing in Cardiology, IEEE, Hangzhou, China, pp. 557–560, 2011.
[5] Center for Machine Learning and Intelligent Systems, “UCI machine learning repository,” Heart Disease
Data Set, 1988. [Online]. Available: https://archive.ics.uci.edu/ml/datasets/heart+disease.
[6] M. P. Alex and S. P. Shaji, “Prediction and diagnosis of heart disease patients using data mining technique,”
in 2019 Int. Conf. on Communication and Signal Processing (ICCSP), Dalian, China, IEEE, pp. 848–852,
2019.
[7] C. B. Gokulnath and S. P. Shantharajah, “An optimized feature selection based on genetic approach and
support vector machine for heart disease,” Cluster Computing, vol. 22, no. S6, pp. 14777–14787, 2018.
[8] A. U. Haq, J. P. Li, M. H. Memon, S. Nazir and R. Sun, “A hybrid intelligent system framework for the
prediction of heart disease using machine learning algorithms,” Mobile Information Systems, vol. 2018, pp.
1–21, 2018.
[9] I. Goodfellow, Y. Bengio and A. Courville, Deep Learning. San Francisco, CA, USA: MIT Press, pp. 1–26,
2016.
[10] W. S. Noble, “What is a support vector machine?,” Nature Biotechnology, vol. 24, no. 12, pp. 1565–1567,
2006.
[11] Y. Shen, Y. Li, H. -T. Zheng, B. Tang and M. Yang, “Enhancing ontology-driven diagnostic reasoning with
a symptom-dependency-aware Naïve Bayes classifier,” BMC Bioinformatics, vol. 20, no. 1, pp. 1–14, 2019.
[12] A. Gupta, L. Kumar, R. Jain and P. Nagrath, “Heart disease prediction using classification (Naive Bayes),”
in Int. Conf. on Computing, Communications, and Cyber-Security, Springer, Singapore, pp. 561–573, 2020.
[13] C. R. Shalizi, “Advanced Data Analysis from an Elementary Point of View,” Pittsburgh, Pennsylvania, USA:
Cambridge University Press, pp. 234–260, 2019. [Online]. Available: https://www.stat.cmu.edu/~cshalizi/
ADAfaEPoV/ADAfaEPoV.pdf.
[14] Scikit-Learn, “Ridge regression and classification, linear models,” 2022. [Online]. Available: https://scikit-
learn.org/stable/modules/linear_model.html#ridge-regression-and-classification.
[15] G. J. McLachlan, “Logistic discrimination,” in Discriminant Analysis and Statistical Pattern Recognition,
1st ed., vol. 1. Queensland, Australia: John Wiley & Sons, pp. 255–282, 2005.
[16] D. J. C. MacKay, “Introduction to Gaussian processes,” Cambridge, United Kingdom, 1998. [Online].
Available: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.81.1927&rep=rep1&type=pdf.
[17] M. A. Friedl and C. E. Brodley, “Decision tree classification of land cover from remotely sensed data,”
Remote Sensing of Environment, vol. 61, no. 3, pp. 399–409, 1997.
[18] A. Liaw and M. Wiener, “Classification and regression by randomForest,” RNews, vol. 2, pp. 18–22, 2002.
[19] S. S. Yadav, S. M. Jadhav, S. Nagrale and N. Patil, “Application of machine learning for the detection of
heart disease,” in ICIMI, Bangalore, India, pp. 165–172, 2020.
[20] L. Medsker and L. C. Jain, “Recurrent neural networks: Design and applications,” in International Series
on Computational Intelligence,1
st ed., vol. 1. Washington D.C., USA: CRC Press, pp. 1–10, 1999.
[21] S. Mohan, C. Thirumalai and G. Srivastava, “Effective heart disease prediction using hybrid machine
learning techniques,” IEEE Access, vol. 7, pp. 81542–81554, 2019.
[22] R. Katarya and S. K. Meena, “Machine learning techniques for heart disease prediction: A comparative
study and analysis,” Health and Technology, vol. 11, no. 1, pp. 87–97, 2020.
CMC, 2023, vol.74, no.3 5883
[23] P. Dileep, K. N. Rao, P. Bodapati, S. Gokuruboyina, R. Peddi et al., “An automatic heart disease prediction
using cluster-based bi-directional LSTM (C-BiLSTM) algorithm,” Neural Computing and Applications,vol.
34, no. 9, pp. 1–14, 2022.
[24] Scikit-Learn, “Chi-squared statistic test feature selection,” 2022. [Online]. Available: https://scikit-learn.
org/stable/modules/generated/sklearn.feature_selection.chi2.html.
[25] M. Ali, “PyCaret: An open source, low-code machine learning library in python,”2020. [Online]. Available:
https://www.pycaret.org.
[26] A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury et al., “PyTorch: An imperative style, high-performance
deep learning library,” in Conf. on Neural Information Processing Systems, Vancouver, Canada, pp. 8024–
8035, 2019.
[27] E. Bisong, “Google Colaboratory,” in Building Machine Learning and Deep Learning Models on Google
Cloud Platform,1
st ed., vol. 1. Berkeley, CA, USA: Apress, pp. 59–64, 2019.
[28] C. R. Harris, K. J. Millman, S. J. Van Der Walt, R. Gommers, P. Virtanen et al., “Array programming with
NumPy,” Nature, vol. 585, no. 7825, pp. 357–362, 2020.
[29] The Pandas Development Team, “Pandas 1.4.2.,” 2022. [Online]. Available: https://pandas.pydata.org/.
[30] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion et al., “Scikit-learn: Machine learning in
python,” Journal of Machine Learning Research, vol. 12, pp. 2825–2830, 2011.
[31] J. D. Hunter, “Matplotlib: A 2D graphics environment,” Computing in Science & Engineering,vol.9,no.3,
pp. 90–95, 2007.
[32] M. Waskom, “Seaborn: Statistical data visualization,” Journal of Open Source Software,vol.6,no.60,
pp. 3021, 2021.
[33] Scikit-Learn, “User guide to feature selection using scikit-learn,” 2022. [Online]. Available: https://scikit-
learn.org/stable/modules/feature_selection.html.
Available via license: CC BY 4.0
Content may be subject to copyright.
