Heart Disease: Application of the K-Nearest Neighbor (KNN) Method

This study employs the frame of mind technique, which is conducted in phases to finish the research. The stages of this research method are shown in Figure 1 and are highly beneficial for researchers in their quest for the truth and a scientific knowledge of phenomena.

1.png

Figure 1. Research methods

2.1 Dataset

A dataset is basically a collection of data. Based on the definition from IBM, a dataset refers to a file that contains one or more records/data. In the data mining process, data alone is not enough. Structured data in the form of datasets is needed so that the results of data analysis will be faster, more accurate, efficient, and of high quality. Therefore, we need data on heart disease (Heart Disease) obtained through Kaggle data. The data will be processed and become an important source of information in this research shown in Figure 2.

2.png

Figure 2. Visualization of datasets

(Source: https://guides.dataverse.org/en/4.6/user/dataset-management.html)

2.2 Preprocessing

Data mining is a sequence of procedures that may be broken down into several phases. Preparing and cleaning the data that is the subject of knowledge discovery in databases must first be cleaned before beginning the data mining process (KDD) [13]. Eliminating redundant data, examining inconsistent data, and fixing data errors such as printing mistakes are all part of the cleaning process [14]. Additionally, a procedure known as enrichment is carried out, which involves adding additional pertinent data or information such as necessary external data or information to the already-existing data in order to improve it for Knowledge Discovery in Databases (KDD). While the concepts in these phases differ, they are still connected to one another. Data mining is one of the stages that makes up the KDD process as a whole.

a) Selection of data before the Knowledge Discovery in Database (KDD) stage of information mining can start, a collection of operational data must be selected.

b) A file kept apart from the working database contains the chosen data that will be utilized for the data mining procedure.

The procedures involved in preprocessing are depicted in Figure 3 and include data cleansing, integration, transformation, reduction, and discretization.

• Data Cleaning: Data cleaning is the process of locating and fixing or eliminating inaccurate data from a dataset. It is also referred to as data cleansing, data scrubbing, or data cleansing. The cleansed data may contain defects that might impede the next steps in the data analysis process, such as being inconsistent, erroneous, duplicate, or mis formatted.
• Data Integration: The practice of merging data from several sources to provide a single, cohesive perspective is known as data integration.
• Data Transformation: The process of transforming data or information from one format to another is another definition of data transformation. For data processing, data analysis, and application development, data transformation is crucial.

3.png

Figure 3. Data preprocessing steps

• Data Reduction: The goal of data reduction is to convert unprocessed data into a more focused, usable, and topic-appropriate format. Statistical tools or categories that facilitate data analysis for researchers might help in data reduction.
• Data Discretization: The process of breaking down a large number of data values into smaller ones to make managing and evaluating the data easier is known as data discretization. Put differently, data discretization refers to a technique that minimizes data loss while transforming the attribute values of continuous data into a limited collection of intervals.

4.1 Dataset

The public dataset from Kaggle was used for this study; it contains text, numbers, photos, and even combinations of other kinds of data. In a specific context, this data may be utilized to find patterns, test hypotheses, and provide answers to problems. The acquired dataset contains information about heart disease. Table 1 has 303 CSV-formatted data about heart illness.

The data on heart disease, which includes age, gender, blood sugar, slope, and chest discomfort, is shown in the table below that was collected from Kaggle in CSV format.

Table 1. Heart disease data

4.2 Preprocessing

Text mining techniques and applications heavily depend on preprocessing procedures. It is the process's initial phase in text mining. The selection data and the selection data findings that will be processed for clustering are the two components of the selection data used in the study by Crone et al. [20].

Table 2 displays the first five rows of the Data Frame data to give an initial view of the dataset structure and counts the number of occurrences of each unique value in the 'target' column and displays the results. This data uses a public dataset on heart disease presented as a CSV file [21].

Table 2. Selection data

To normalize the data is done using Min-Max scaling; with the formula as given below, this will scale all numerical values in the range of 0 to 1.

Figure 4 provides an overview of the class distribution in the target column. Created a bar plot using Seaborn to display the class distribution in the 'target' column. On the x-axis, it has 'target'. This figure also presents the amount of data according to gender, where 0 is data on female gender which is obtained as much as 138 and 1 presents data with male gender as much as 165. In this data, it can be concluded that there are more males than females. 

${{\text{X}}_{\text{normalized }}}=\frac{\left( \text{X}-{{\text{X}}_{\min }} \right)}{\left( {{\text{X}}_{\max }}-{{\text{X}}_{\min }} \right)}$

4.png

Figure 4. Result of selection data based on target

4.3 Clustering

Clustering or classification is a method used to divide a data set into groups based on predefined similarities [21]. This research uses the K-Nearest Neighbor Algorithm to classify the risk level of heart disease based on age and gender. Data sources as objects in this study is data taken from the Kaggle.com website. The data used in This study consists of attributes or variables such as gender, age, slope, blood sugar, and chest pain.

Figure 5 visualizes the distribution of count data based on the 'sex' column. This count plot provides information about the number of observations for each 'sex' category (0 for female, 1 for male). The plot does not display a legend and uses the 'mako_r' color palette. The x-axis shows the 'sex' category, and the y-axis shows the number of observations for each category. The program helps provide an initial understanding of the distribution of the 'sex' variable in the dataset.

5.png

Figure 5. Count sex

Figure 6 provides useful information about the gender distribution, summary statistics related to the target, and visualization of the frequency of heart disease by age in the dataset. The figure displays a graph where the largest value of heart disease is in women based on age 58.

6.png

Figure 6. Frequency for age

In Figure 7, this scatter plot helps visualize the distribution of age and maximum heart rate data for both patient groups. With the red and blue colors and the added legend, this plot makes it easier to interpret the data and understand possible patterns in the relationship between age and maximum heart rate based on heart disease status. The blue color (Not Disease) shows more of the plot.

7.png

Figure 7. Maximum heart rate maximum heart rate for age

The bar chart in Figure 8 shows the frequency of occurrence of combinations of values from the 'slope' and 'target' columns, with different colors for each target value. The relationship between the variable 'slope' (slope of the peak exercise ST segment) and the target variable ('target' which may indicate the presence or absence of heart disease). Through the bar chart, it is possible to see how often combinations of slope and target values occur, as well as a comparison of the frequency between different target values for each slope value. This helps in understanding the pattern of relationship between the two in understanding the pattern of relationship between the two variables.

8.png

Figure 8. Heart disease frequency for slope

The bar chart in Figure 9 visualizes the frequency of heart disease by Fasting Blood Sugar (FBS). It also clearly shows the distribution of the presence or absence of heart disease in groups with high and low FBS values. The differentiated colors make it easy to identify patterns, with the first color indicating those without heart disease and the second color indicating those with heart disease.

9.png

Figure 9. Heart disease frequency according to FBS

Figure 10 visualizes the frequency of presence or absence of heart disease based on chest pain type. Through this diagram, we can see the distribution of the number of heart disease cases for each type of chest pain. Different colors are used to distinguish between those with heart disease (first color) and those without heart disease (second color).

10.png

Figure 10. Heart disease frequency according to chest pain type

This program converts the categorical columns in the dataframe into dummy variables using pd.get_dummies (Figure 11), which converts each unique category into a column that is aligned with the current awalan. After transforming the columns "cp," "thal," and "slope" into dummies, the results are merged back into the original dataframe. This step aims to smooth the data so that it is more compatible with machine learning algorithms, which often need numerical input. Essentially, this tool preprocesses data to convert categorical columns into a format that may be used for further in-depth analysis or machine learning models.

11.png

Figure 11. Frames data categories 1

The next step is to delete the original columns "cp," "thal," and "slope" from the dataframe data using data.drop(columns = ['cp', 'thal','slope']) (Figure 12). This is done after transforming the categorical columns "cp," "thal," and "slope" into dummy variables and merging them back with the original dataframe. This guarantees that the dataframe keeps only the created fake columns and the remaining, unaltered columns. Because the initial category columns are no longer redundant, the resulting dataframe is clearer and prepared for machine learning analysis or modeling. To confirm that the column deletion was effective and the intended data structure was achieved, the final result displays the first five rows of the dataframe that have been processed.

12.png

Figure 12. Frames data categories 2

In Figure 13 showing the visualization results, the program looks for the K value that gives the highest accuracy on the test data. The maximum accuracy result along with the corresponding K value is then printed. Overall, this graph helps in determining the optimal parameters (number of neighbors) for the KNN model being built, with the aim of improving the performance and generalization of the model on data that has never been seen before [22]. This graph is the end result of several previous steps by displaying this plot can select the k value that gives the highest accuracy score to the KNN model, helping in decision making related to hyperparameter settings.

13.png

Figure 13. K-Nearest Neighbor (KNN) value