Digitalization in medicine has become one of the largest gaps in almost all healthcare systems in the world. Diabetes remains one of the pressing health problems. According to World Health Organization, the number of people with diabetes increased from 108 million in 1980 to 422 million in 2014. This research is devoted to solving the problem of classifying patients with diabetes and diagnosing this disease. To solve the problem, a machine learning model was built based on a decision tree method. To develop the model, an open database of patients with diabetes, consisting of 768 patients, was used. On the foundation of the constructed model, a software package in the Python language has been developed.
solve medical tasks [
The world pandemic of the new coronavirus has changed the usual way of life and approaches to
Right now, digitalization in medicine has become one of the largest gaps in almost all healthcare
systems in the world. As practice shows, digital technologies can significantly improve the quality of
healthcare [
Diabetes remains one of the pressing health problems. According to World Health Organization, the number of people with diabetes increased from 108 million in 1980 to 422 million in 2014 [24]. The global prevalence of diabetes among people over 18 years of age increased from 4.7% in 1980 to 8.5% in 2014 [25]. Premature mortality from diabetes increased by 5% between 2000 and 2016 [26]. Some scientists have linked cases of childhood diabetes with COVID-19 [27].
Diabetes mellitus is a chronic disease of the endocrine system, which is caused by a violation of insulin synthesis and an increase in blood sugar. The disease can lead to the development of a number of serious deficiencies. There are 2 main types of diabetes mellitus: types I and II, as well as gestational diabetes in pregnant women and symptomatic diabetes. Let's consider 2 main types, from which more and more people are suffering from all over the globe every day. Diabetes mellitus type I is more EMAIL: (T. Dudkina); (I. Meniailov);
2021 Copyright for this paper by its authors. common in patients under the age of 30, more often it develops due to the fact that the pancreas begins to work worse against the background of a viral infection or the action of toxins. With this type of diabetes, the body is unable to produce insulin, thus, when diagnosed with type I diabetes, the patient becomes insulin dependent throughout his life. Diabetes mellitus type II - occurs due to insulin resistance, of which they get sick more often. Older people are more prone to it, because sugar tolerance decreases over the years. There are a number of factors that increase your risk of developing type II diabetes. [28].
The symptoms depend on how long the person is sick, on the severity of the disease and the patient's personal immunity. Someone may have a vivid clinical picture right away, while someone may have a barely noticeable clinic or, even worse, be absent. Diabetes can be diagnosed using a variety of diagnostics. The main method for diagnosing diabetes mellitus is laboratory tests of urine and blood for glucose levels. In some cases, the doctor may prescribe an ultrasound of the kidneys, an EEG of the brain, etc. People at risk should carefully monitor their blood glucose and blood pressure levels.
An important challenge in the fight against diabetes is the classification of patients and the diagnosis of the disease. To solve this problem, it is advisable to use a machine learning apparatus. In modern science, several models have been implemented that make it possible to diagnose diabetes according to specified parameters.
Sisodia S. and Sisodia D.S. have made prediction model of Diabetes using naïve Bayes algorithm with accuracy 76.3% [29]. Naveen K. has made classification model of Diabetes using SVM algorithm and data of glucose and blood pressure [30].
These and other analyzed researches show the limitations of the factors used to train the model, which leads to a decrease in the classification accuracy.
The aim of the research is to build a model that allows classifying a person's condition in relation to the incidence of diabetes using machine learning methods. 2.
To solve the problem, we use the decision trees method [31-32]. The decision tree is a sequential hierarchical structure and includes: branches with attributes on which the result depends - the objective function; nodes - random vertices in which possible scenarios for the development of events are determined; leaf (leaf) nodes with objective function values represent the final results of choosing a specific attribute value and combine several objects. Decision trees are divided into two types by the type of predicted indicator: classification trees and regression trees. When developing a system for establishing a diagnosis, it is advisable to use classification trees, since they are used research on certain attributes, namely, to attribute objects (symptoms) from a previously known class (a certain disease). Decision trees divide data into groups, resulting in a hierarchy of "if ... then ..." operators that classifies data. partition where we maximize the increment is:
Let's define an objective function in order to divide the nodes into informative functions. Each ( , ) = ( ) − ∑ =1 ( ) samples in the j-th child node. our case, child nodes Dleft and Dright are: where f is attribute by which splitting is performed; Dp and Dj are parent and j-th child nodes; Ι is a measure of heterogeneity; Np is the total number of samples in the parent node; Nj is the number of
For simplicity and to reduce the combinatorial search space, we implement binary decision trees. In where f is attribute by which splitting is performed; Dp and Dj are datasets of parent and j-th child nodes; Ι is a measure of heterogeneity; Np is total number of samples in parent node; Dleft and Dright are child ( , ) = ( ) − (
) −
ℎ ( ℎ ) (1) (2) nodes; Nleft and Nright are numbers of patterns in left and right child nodes; Nj is number of samples in j-th child node.
Determination of entropy for all non-empty classes ( | ) ≠ 02: we have a uniform distribution of classes. of misclassification: where ( | ) is fraction of samples that belongs to class and single node t.
So, the entropy is 0 if all samples in a node belong to the same class, and the entropy is maximal if The Gini measure of heterogeneity [33] can be perceived as a criterion that minimizes the likelihood ( ) = − ∑ =1 ( | ) 2 ( | ) (3) (4) (5)
( ) = ∑ =1 ( | )(1 − ( | )) = 1 − ∑ =1 ( | )2 where ( | ) is fraction of samples that belongs to a class and a single node t; LG(t) is Gini measure of heterogeneity.
Another measure of heterogeneity is classification error:
( ) = 1 − max { ( | )} where ( | ) is fraction of samples that belongs to a class and a single node t; ( ) is classification error.
This criterion is suitable for tree pruning, but is not recommended for tree growth because it is less sensitive to changes in the capabilities of the classes in the nodes.
In order to build a decision tree, you need certain data. The Pima Indians Diabetes DataBase was used to test the diabetes diagnostic model. Database has 768 instances and 9 attributes for individual patients (Table 1). will fall to him if he first learns these signs from a doctor.
As we can see from the plot, there are some outliers in some of the columns.
We can see that there are 0 values for blood pressure. So, we assume that it is mistake data. Observing the data, we see 35 samples, where the value is 0. Even after fasting, your glucose level will not go below zero. Therefore, zero is a misread. Before further use of the selection, remove the lines with “BloodPressure”, “BMI” and “Glucose” equal to zero.
The Spyder development environment was used to write the code to build the decision tree. In order to read our data from the table, the Pandas library was used. The Scikit-learn machine learning library was also used. To implement a decision tree, you need to import the required Python packages. Then we upload our database.
The next step is to split this data into two parts - training data and testing data. Next, you need to train the model using the DecisionTreeClassifier class (Scikit-learn library). Next, we make a forecast, and we also need to get an accuracy estimate, a classification report and an error matrix. The final step is to render our decision tree [34]. The color of the nodes is used to highlight the class that has the most in each node and to convey the names of the classes and traits so that the tree is correctly marked up.
For the first experiment, the data was split as follows: 70% for training and 30% for testing. The results are shown in Figures 2 and 3.
For the second experiment, the data was split as follows: 50% for training and 50% for testing. The results are shown in Figures 4 and 5.
For the third experiment, the data was split as follows: 30% for training and 70% for testing. The results are shown in Figures 6 and 7.
Overall, it can be said that decision tree analysis is a predictive modeling tool that can be applied in many areas. Decision trees can be built using an algorithmic approach that can partition the dataset in different ways depending on conditions.
After the work done, we can conclude that the more data is allocated for training the model, the better the accuracy estimate we get. In our case, the best option is to split the data by 50% for training the model and 50% for testing, since the accuracy of this option is 0.71.
After analyzing the constructed diagnostic model, the following advantages can be identified: fast learning process; generation of rules in areas where it is difficult for an expert to formalize his knowledge; intuitive classification model; high prediction accuracy, comparable to other methods of data analysis (statistics, neural networks); construction of nonparametric models.
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” [35]. [20] V. P. Mashtalir, et al. Group structures on quotient sets in classification problems, Cybernetics and
Systems Analysis 50 (4) (2014) 507-518. [21] I. Meniailov, et. al. Using the K-means method for diagnosing cancer stage using the Pandas library, CEUR, 2386 (2019) 107-116. [22] D. Chumachenko, et. al. On agent-based approach to influenza and acute respiratory virus infection simulation, 14th International Conference on Advanced Trends in Radioelectronics, Telecommunications and Computer Engineering (2018) 192-196. doi: 10.1109/TCSET.2018.8336184 [23] K. Bazilevych, et.al. Determining the Probability of Heart Disease using Data Mining Methods.
CEUR, 2488 (2019) 1-12. [24] G. Valenti, G, Tamma, History of Diabetes Insipidus, Giornale italiano di nefrologia 33 (2016) 66:33.S66.1. [25] N. Sarwar, et al. Diabetes mellitus, fasting blood glucose concentration, and risk of vascular disease: a collaborative meta-analysis of 102 prospective studies, Lancet 375 (9733) (2014) 22152222. [26] D. Chumachenko, et. al. On Intelligent Decision Making in Multiagent Systems in Conditions of Uncertainty, Proceedings of 2019 11th International Scientific and Practical Conference on Electronics and Information Technologies (2019) 150-154. doi: 10.1109/ELIT.2019.8892307 [27] A. Hussain, B. Bhowmik, N. C. do Vale Moreira, COVID-19 and diabetes: Knowledge in progress, Diabetes Research and Clinical Practice 162 (2020) 108142. [28] D.G. Bichet, Genetics and diagnosis of central diabetes insipidus, Annales d'Endocrinologie 73 (2) (2012) 117-127. [29] V. Yesina, et. al., Method of Data Openness Estimation Based on User-Experience in Infocommunication Systems of Municipal Enterprises, International Scientific-Practical Conference on Problems of Infocommunications Science and Technology (2019) 171–176. doi: 10.1109/INFOCOMMST.2018.8631897 [30] K.G. Naveen, et. al., Prediction of diabetes using Machine Learning classification algorithms,
International journal of scientific and technology research 9 (1) (2020) 1805-1808. [31] A. Albu, From logical inference to decision trees in medical diagnosis, Proceedings of 2017 E
Health and Bioengineering Conference (2017) 65-68. doi: 10.1109/EHB.2017.7995362 [32] M.D.A. Praveena, J. S. Krupa, S. SaiPreethi, Statistical Analysis Of Medical Appointments Using Decision Tree, Conference on Science Technology Engineering and Mathematics (2019) 59-64. doi: 10.1109/ICONSTEM.2019.8918766 [33] D. Chumachenko, O. Sokolov, S. Yakovlev, Fuzzy recurrent mappings in multiagent simulation of population dynamics, International Journal of Computing 19 (2) (2020) 290-297. [34] S. N. Gerasin, et. al., Set coverings and tolerance relations, Cybernetics and Systems Analysis 44 (3) (2008) 333-340. [35] Yakovlev S. , et. al., The concept of developing a decision support system for the epidemic morbidity control, CEUR, 2753 (2020) 265–274.