=Paper=
{{Paper
|id=Vol-3641/short3
|storemode=property
|title=Classification of Patients with Suspected Coronary Artery Disease Based on Locally Weighted Least Squares Method
|pdfUrl=https://ceur-ws.org/Vol-3641/short3.pdf
|volume=Vol-3641
|authors=Mykola Butkevych,Kseniia Bazilevych,Iryna Trofymova
|dblpUrl=https://dblp.org/rec/conf/profitai/ButkevychBT23
}}
==Classification of Patients with Suspected Coronary Artery Disease Based on Locally Weighted Least Squares Method==
https://ceur-ws.org/Vol-3641/short3.pdf
Classification of Patients with Suspected Coronary Artery
Disease Based on Locally Weighted Least Squares Method
Mykola Butkevych, Kseniia Bazilevych and Iryna Trofymova
National Aerospace University “Kharkiv Aviation Institute”, Chkalow str. 17, Kharkiv, 61070, Ukraine
Abstract
The study explores the application of a linear regression model integrated with Locally Weighted Least
Squares for diagnosing coronary artery disease. We utilized a well-established dataset, applying the
locally weighted least squares method to enhance the model's sensitivity to local data variations. Our
results yielded a Mean Squared Error of 0.1282 and a Mean Absolute Error of 0.2316, indicating a model
with reasonable predictive accuracy. The study concludes that while the approach shows promise, it
necessitates further refinement and exploration with more diverse datasets and advanced techniques.
This research contributes to the evolving landscape of coronary artery disease diagnostics, aiming to
improve prediction accuracy and patient outcomes.
Keywords
Coronary artery disease, machine learning, data-driven medicine, diagnostics, classification 1
1. Introduction
Cardiovascular diseases (CVDs) remain a significant public health challenge globally,
representing the leading cause of mortality and contributing substantially to healthcare burdens
[1]. Among various CVDs, coronary artery disease (CAD) is particularly noteworthy due to its high
prevalence and potential to lead to severe outcomes such as heart attacks and heart failure [2].
CAD arises primarily from the buildup of plaques within coronary arteries, leading to impaired
blood flow to the heart muscle [3]. The early detection and management of CAD are crucial for
improving patient outcomes and reducing the risk of life-threatening events. However, accurately
diagnosing CAD in its early stages remains challenging, often requiring a combination of clinical
evaluation, imaging tests, and invasive procedures [4].
The advent of information technologies has brought transformative changes to the healthcare
sector. Digital tools and platforms have enhanced the efficiency and quality of care delivery,
enabling healthcare providers to effectively manage and analyze large volumes of patient data
[5]. Integrating electronic health records, telemedicine, and mobile health applications has
improved access to healthcare services and facilitated more personalized and patient-centered
care [6]. Information technologies have also played a pivotal role in advancing medical research
and development, driving innovations in diagnostic techniques, treatment strategies, and disease
management [7].
Data-driven diagnostics represent a paradigm shift in disease detection and management [8].
Leveraging the power of big data analytics, machine learning, and artificial intelligence, this
approach emphasizes utilizing vast and diverse healthcare data to uncover patterns, make
predictions, and aid in clinical decision-making [9]. By analyzing data from sources such as
electronic health records, imaging studies, and genomic profiles, data-driven diagnostics can offer
insights beyond traditional diagnostic methods. This method promises to enhance the accuracy,
ProfIT AI 2023: 3rd International Workshop of IT-professionals on Artificial Intelligence (ProfIT AI 2023), November
20–22, 2023, Waterloo, Canada
nikolai.butkevych@gmail.com (M. Butkevych); ksenia.bazilevych@gmail.com (K. Bazilevych);
irina.trofymova@gmail.com (I. Trofymova)
0000-0001-8189-631x (M. Butkevych); 0000-0001-5332-9545 (K. Bazilevych); 0000-0002-1537-5601
(I. Trofymova)
© 2023 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
efficiency, and personalization of diagnostics, particularly for complex diseases where
conventional approaches may fall short.
Applying modeling techniques, such as the Locally Weighted Least Squares Method, to
detecting suspected CAD significantly advances cardiovascular diagnostics [10]. This approach
entails creating a predictive model that can classify patients based on the likelihood of having
CAD, utilizing various clinical and demographic variables. The locally weighted least squares
method, known for its ability to model complex, nonlinear relationships in data, is particularly
suited for handling the multifaceted nature of CAD. Applying this method makes it possible to
identify subtle patterns and associations in patient data that might indicate the presence of CAD,
thereby aiding in early detection and timely intervention. This novel approach underscores the
potential of combining advanced analytical techniques with clinical expertise to improve the
detection and management of coronary artery disease.
The aim of the paper is to develop the model of classification of patients with suspected
coronary artery disease using locally weighted least squares method.
2. Current research analysis
The current research landscape in the domain of CAD diagnosis has increasingly gravitated
towards integrating advanced analytical techniques with traditional clinical methods. This
synergy is driven by the growing recognition that conventional diagnostic approaches, while
effective, may only partially capture the complexity and heterogeneity inherent in CAD. Recent
studies have demonstrated a keen interest in exploring data-driven models, particularly those
employing machine learning and statistical analysis, to enhance the precision and predictive
power of CAD diagnostics. These investigations typically focus on using patient data,
encompassing clinical parameters, imaging results, and biochemical markers, to develop
algorithms capable of identifying patterns indicative of CAD with greater accuracy and efficiency
than traditional methods alone. The thrust of this research underscores a paradigm shift towards
more personalized and preemptive healthcare strategies, underlining the significance of early
and accurate detection of CAD for better patient outcomes.
The research [11] focuses on evaluating a hybrid system combining Genetic Algorithms (GAs),
Biogeography-Based Optimization (BBO), and Particle Swarm Optimization (PSO) with neural
networks for heart disease diagnosis. The study employs the Z-Alizadeh Sani dataset, which
contains 303 records. The model utilizes the top 14 weight features from the dataset, determined
through trial and error, noting that increasing the number of features did not enhance prediction
accuracy. The performance of the developed models, GAsBBO-MLPNNs, BBO-MLPNNs, and PSO-
MLPNNs, is contingent on several factors, including the number of iterations, the number of
neurons in hidden layers, population size, and the activation function of the hidden layers. The
research employs a tenfold cross-validation method for evaluating the models, using 90 percent
of the dataset for training and the remaining 10 percent for testing. Performance metrics such as
overall accuracy, F-score, confusion matrix, Sensitivity (Recall), and Specificity are used to assess
the effectiveness of the proposed models.
The paper [12] introduces a novel wrapper feature selection method for diagnosing coronary
artery disease, utilizing Grey Wolf Optimization (GWO) and Support Vector Machine (SVM)
classifier. The study proposes a two-stage approach to address the challenge of large datasets in
medical diagnostics, which often contain redundant and irrelevant features. Initially, GWO is
employed for efficient feature selection in the disease identification dataset, aiming to enhance
the relevance and quality of the data used. Subsequently, the fitness function of GWO is evaluated
using an SVM classifier. The methodology is validated using the Cleveland Heart disease dataset,
with the results demonstrating that the proposed GWO-SVM method surpasses current
approaches, achieving 89.83% accuracy, 93% sensitivity, and 91% specificity. This research
highlights the potential of integrating advanced optimization techniques with machine learning
classifiers to improve critical illnesses like coronary artery disease diagnosis.
The paper [13] proposes a new classification system for coronary artery abnormalities (CAAs)
following Kawasaki Disease, using only coronary artery z-scores. This system was developed
after reviewing echocardiograms from 1990 to 2007 of patients with a history of Kawasaki
Disease. The study focuses on refining the classification of CAAs, arguing that current methods
underestimate their severity. By analyzing z-scores and their distribution, the study suggests an
optimized definition of CAA sizes. This research is pivotal in accurately identifying and classifying
CAAs in Kawasaki Disease, which is crucial for effective management and prognosis.
The study [14] employs Chi-squared Automatic Interaction Detection (CHAID) to explore the
interrelation of significant risk factors in the development of CAD. It includes a retrospective
analysis of 1381 patients who underwent coronary angiography at a cardiology clinic between
January 1999 and February 2003. The research assesses various factors such as sex, age, diabetes,
hypercholesterolemia, hypertension, smoking status, family history of CAD, and body mass index.
The findings highlight sex and age as primary risk factors, with diabetes mellitus being notably
significant in certain age groups. For older females, hypercholesterolemia emerges as a key
predictor. The study concludes by ranking the risk factors in order of their classification
importance for CAD.
The paper [15] presents a novel method for assessing the severity of CAD from electronic
health records. It utilizes a recurrent capsule network model to extract semantic relations from
coronary arteriography texts, primarily using Chinese datasets. The model's performance is
validated on data collected from Shanghai Shuguang Hospital, showcasing high accuracy and
efficiency in CAD severity classification. This approach represents a significant advancement in
using deep learning techniques for the automated analysis of CAD severity, demonstrating the
potential for improved diagnostic methods in healthcare.
The analyzed papers collectively illustrate the dynamic and innovative landscape of research
in CAD diagnostics. They underscore a shift towards integrating advanced computational
methods, such as machine learning, neural networks, and optimization algorithms, with
traditional clinical approaches. This convergence has led to more accurate and efficient diagnostic
tools, reflecting a trend toward personalized medicine and improved patient outcomes. The
studies, ranging from feature selection methodologies to severity classification models,
demonstrate the potential of these advanced techniques in enhancing the precision of CAD
diagnostics. This evolving field continues to offer promising avenues for research, potentially
impacting the future of cardiovascular healthcare significantly.
3. Materials and methods
In the field of medical diagnostics, the classification of diseases such as CAD is a crucial task [16].
Our methodological approach in this study employs the Locally Weighted Least Squares (LWLS)
technique to train a linear regression model specifically for CAD diagnosis. This involves selecting
a representative patient dataset, applying LWLS to train the model, and using this model to
predict CAD in new patient data. The performance of this model is critically evaluated against
known outcomes to ensure its accuracy and reliability, with iterative improvements made based
on these assessments. This approach aims to offer a nuanced and effective tool for CAD diagnosis.
Linear regression is a statistical approach to modeling the relationship between a dependent
variable and one or more independent variables [17]. The dependent variable in the context of
CAD diagnosis is the likelihood of CAD presence. In contrast, the independent variables are
various patient data points, such as age, cholesterol levels, and blood pressure. The general form
of the model is:
𝑦 = 𝛽0 + 𝛽1 𝑥1 + 𝛽2 𝑥2 + ⋯ + 𝛽𝑛 𝑥𝑛 + 𝜖, (1)
where y represents the predicted outcome (e.g., the probability of CAD); 0 is the y-intercept, 1,
2, …, n are coefficients that represent the impact of each independent variable x1, x2, …, xn on the
predicted outcome, and 𝜖 is the error term, accounting for the deviation of the predictions from
the actual values.
LWLS is an enhancement of linear regression, which applies a weighting scheme to the data
points [18]. Each data point gets a weight based on its proximity to the point where the prediction
is made. The closer the data point to the prediction point, the higher its weight. This is expressed
mathematically as:
𝑛
2
(2)
min ∑ 𝜔𝑖 (𝑦𝑖 − (𝛽0 + 𝛽1 𝑥𝑖1 + ⋯ + 𝛽𝑛 𝑥𝑖𝑛 )) ,
𝛽
𝑖=1
where i is is the weight assigned to the i-th data point. The weights are typically assigned based
on a function of the distance between the data points and the point of prediction, with common
choices being Gaussian or exponential functions.
The integration of linear regression with LWLS for CAD diagnosis involves using the weighted
least squares approach within the linear regression framework. This combination tailors the
model to local data variations, making it more adaptable and sensitive to the specific
characteristics of the CAD dataset. Doing so aims to increase the accuracy of CAD predictions,
acknowledging that the importance and influence of specific risk factors can vary across different
patient groups.
For our study we have used the Framingham Heart Study dataset publicly available on Kaggle
[19]. The dataset is a collection of data from the renowned Framingham Heart Study focused on
identifying common factors or characteristics contributing to cardiovascular disease. The dataset
contains various patient information, such as age, sex, blood pressure, cholesterol levels, and
smoking status. It is widely used in medical research, particularly in studies related to heart
disease and its risk factors. This dataset is valuable for developing predictive models and
conducting epidemiological studies in cardiovascular health. The characteristic of the dataset is
presented in Table 1.
Table 1
The dataset characteristic
Attribute Scale type Range
Sex Boolean 0, 1
Age Metric 32..70
Education Metric 1..4
CurrentSmoker Boolean 0, 1
CigsPerDay Metric 0..43
BPMeds Boolean 0, 1
PrevalentStroke Boolean 0, 1
PrevalentHyp Boolean 0, 1
Diabetes Boolean 0, 1
TotChol Metric 143..696
SysBP Metric 83,5..295
DiaBP Metric 48..143
BMI Metric 15,54..38,53
HeartRate Metric 50..110
Glucose Metric 45..268
TenYearCHD Boolean 0, 1
4. Results
The training process of a model is a meticulous and multifaceted procedure. It starts with data
preparation, which involves thoroughly cleaning and organizing the dataset, addressing any
missing values, and standardizing variables on different scales. Special attention is given to the
nature of each variable, as the dataset comprises a blend of categorical (Boolean) and continuous
(Metric) variables.
Selecting the most relevant features is a critical step that impacts the model's effectiveness.
This selection is based on the characteristics of the data and the goal of predicting the
TenYearCHD outcome.
A logistic regression model is typically chosen for its appropriateness in handling binary
outcomes and its ability to provide probabilities. The training phase involves fitting this model to
the selected features using a designated portion of the dataset.
Following training, the model undergoes a rigorous evaluation to assess its accuracy,
precision, recall, and other relevant metrics. This evaluation is crucial in determining the model's
effectiveness in predicting coronary heart disease.
If the initial results are unsatisfactory, the model undergoes tuning, where hyperparameters
are adjusted to optimize performance. Cross-validation techniques are employed to ensure the
model's robustness and generalizability.
This process is inherently iterative, requiring continuous refinement and adjustments based
on the performance metrics and validation outcomes. This rigorous methodology ensures the
development of a reliable and accurate predictive model.
The results are presented in Table 2.
Table 2
Modeling results
Metric Value
Mean Squared Error 0.1282
Mean Absolute Error 0.2316
The results indicating a Mean Squared Error (MSE) of 0.1282 and a Mean Absolute Error
(MAE) of 0.2316 are pretty revealing. The MSE, being relatively low, suggests that the model's
predictions are generally close to the actual values. This is indicative of a model with a good fit to
the data. However, it is essential to consider the complexity of the dataset and the nature of CAD
prediction, where even small errors can be clinically significant.
The MAE provides a straightforward interpretation of the average magnitude of prediction
errors without considering their direction. An MAE of 0.2316, in the context of CAD prediction,
can be considered moderate. It highlights that while the model has predictive validity, there is
still room for improvement, especially in reducing false positives and false negatives, which are
critical in medical diagnostics.
These results should be contextualized within the study's limitations, including the dataset's
representativeness and the model's generalizability to other populations. Future work could
focus on incorporating more diverse datasets, exploring more complex models, or integrating
additional relevant features that could enhance the model's predictive accuracy. Additionally, the
implications of these findings for clinical practice should be cautiously interpreted, considering
the balance between the benefits of early detection and the risks of over-diagnosis.
5. Conclusions
In the conclusion of this study, we underscore the criticality of enhancing CAD diagnostics in light
of its increasing prevalence. The methodological novelty introduced by combining linear
regression with Locally Weighted Least Squares has shown significant promise. Our findings,
reflecting a balance of accuracy and areas for improvement, indicate a substantial stride in CAD
diagnostic approaches. Future research directions aim to enrich the model's robustness through
diverse and comprehensive datasets and explore the integration of more sophisticated
techniques, possibly encompassing artificial intelligence. Such advancements could revolutionize
CAD diagnostics, contributing immensely to predictive medicine and improving patient outcomes
in cardiovascular health. This research serves as a stepping stone towards more accurate,
efficient, and non-invasive diagnostic methods, aligning with the goal of enhancing healthcare
delivery and patient care in cardiovascular diseases.
Acknowledgements
The study was funded by the National Research Foundation of Ukraine in the framework of the
research project 2020.02/0404 on the topic “Development of intelligent technologies for
assessing the epidemic situation to support decision-making within the population biosafety
management”.
References
[1] E. Goldsborough, N. Osuji, and M. J. Blaha, “Assessment of Cardiovascular Disease Risk: A 2022
Update,” Endocrinology and Metabolism Clinics, vol. 51, no. 3, pp. 483–509, Sep. 2022, doi:
10.1016/j.ecl.2022.02.005.
[2] A. Kr. Malakar, D. Choudhury, B. Halder, P. Paul, A. Uddin, and S. Chakraborty, “A Review on
Coronary Artery disease, Its Risk factors, and Therapeutics,” Journal of Cellular Physiology, vol.
234, no. 10, pp. 16812–16823, Feb. 2019, doi: 10.1002/jcp.28350.
[3] A. Lala and A. S. Desai, “The Role of Coronary Artery Disease in Heart Failure,” Heart Failure
Clinics, vol. 10, no. 2, pp. 353–365, Apr. 2014, doi: 10.1016/j.hfc.2013.10.002.
[4] P. G. Steg and G. Ducrocq, “Future of the Prevention and Treatment of Coronary Artery
Disease,” Circulation Journal: Official Journal of the Japanese Circulation Society, vol. 80, no. 5, pp.
1067–1072, Apr. 2016, doi: 10.1253/circj.CJ-16-0266.
[5] M. Mazorchuck, et al., “Web-Application Development for Tasks of Prediction in Medical
Domain,” 2018 IEEE 13th International Scientific and Technical Conference on Computer Sciences
and Information Technologies (CSIT), pp. 5–7, Sep. 2018, doi: 10.1109/stc-csit.2018.8526684.
[6] K. Bazilevych, M. Butkevych, and N. Dotsenko, “Cardiac Studies Diagnostic Data Informative
Features Investigation based on Cumulative Frequency Analysis,” CEUR Workshop Proceedings,
vol. 3348, pp. 84–89, 2022.
[7] D. Chumachenko, “On Intelligent Multiagent Approach to Viral Hepatitis B Epidemic Processes
Simulation,” 2018 IEEE Second International Conference on Data Stream Mining & Processing
(DSMP), pp. 415–419, Aug. 2018, doi: 10.1109/dsmp.2018.8478602.
[8] N. Dotsenko, et al., “Modeling of the Processes of Stakeholder Involvement in Command
Management in a Multi-Project Environment,” 2018 IEEE 13th International Scientific and
Technical Conference on Computer Sciences and Information Technologies (CSIT), pp. 29–32, Sep.
2018, doi: 10.1109/stc-csit.2018.8526613.
[9] S. Yakovlev et al., “The Concept of Developing a Decision Support System for the Epidemic
Morbidity Control,” CEUR Workshop Proceedings, vol. 2753, pp. 265–274, 2020.
[10] M. Bevilacqua and F. Marini, “Local classification: Locally weighted–partial least squares-
discriminant analysis (LW–PLS-DA),” Analytica Chimica Acta, vol. 838, pp. 20–30, Aug. 2014, doi:
10.1016/j.aca.2014.05.057.
[11] M. I. Dwaikat and M. Awad, “Hybrid Model for Coronary Artery Disease Classification Based
on Neural Networks and Evolutionary Algorithms,” Journal of Information Science and
Engineering, vol. 38, no. 5, pp. 1001–1020, 2022.
[12] Q. Al-Tashi, H. Rais, and S. Jadid, “Feature Selection Method Based on Grey Wolf Optimization
for Coronary Artery Disease Classification,” Advances in intelligent systems and computing, vol.
843, pp. 257–266, Sep. 2018, doi: 10.1007/978-3-319-99007-1_25.
[13] C. Manlhiot, K. Millar, F. Golding, and B. W. McCrindle, “Improved Classification of Coronary
Artery Abnormalities Based Only on Coronary Artery z-Scores After Kawasaki Disease,” Pediatric
Cardiology, vol. 31, no. 2, pp. 242–249, Dec. 2009, doi: 10.1007/s00246-009-9599-7.
[14] T. Mevlüt, K. Imran, and K. Turhan, “Analysis of intervariable relationships between major
risk factors in the development of coronary artery disease: A classification tree approach,”
Anadolu Kardiyoloji Dergisi, vol. 7, no. 2, pp. 140–145, 2007.
[15] Q. Wang, J. Qiu, Y. Zhou, T. Ruan, D. Gao, and J. Gao, “Automatic Severity Classification of
Coronary Artery Disease via Recurrent Capsule Network,” Proceedings - 2018 IEEE International
Conference on Bioinformatics and Biomedicine, BIBM 2018, pp. 1587–1594, Dec. 2019, doi:
10.1109/bibm.2018.8621136.
[16] V. P. Mashtalir, V. V. Shlyakhov, and S. V. Yakovlev, “Group Structures on Quotient Sets in
Classification Problems,” Cybernetics and Systems Analysis, vol. 50, no. 4, pp. 507–518, Jul. 2014,
doi: 10.1007/s10559-014-9639-z.
[17] T. Masuda et al., “Development and Validation of Generalized Linear Regression Models to
Predict Vessel Enhancement on Coronary CT Angiography,” Korean Journal of Radiology, vol. 19,
no. 6, pp. 1021–1021, Jan. 2018, doi: 10.3348/kjr.2018.19.6.1021.
[18] Y. Xie, “Fault monitoring based on locally weighted probabilistic kernel partial least square
for nonlinear time‐varying processes,” Journal of Chemometrics, vol. 33, no. 12, Dec. 2019, doi:
10.1002/cem.3196.
[19] A. Bhardwaj, “Framingham heart disease dataset,” Kaggle, 2021.
https://www.kaggle.com/datasets/aasheesh200/framingham-heart-study-dataset (accessed
Sep. 01, 2023).