Hybrid Ensemble Model with Bagging (HEMBAGG) Leveraging Machine Learning for Breast Carcinoma Diagnosis

It is among the most widespread kinds of cancer and a substantial global impact on female mortality worldwide [1]. It is a general issue instead of a regional one. It is usually considered to be higher than fifty percent in most countries. The type with the greatest incidence is breast cancer in comparison to other cancer types like liver cancer, cervical cancer, lung cancer, etc. More than three million females are estimated to be affected annually in India. Although breast cancer has no proven medication then in this scenario early detection and diagnosis may significantly enhance survival rates of this among females. The lack of clear symptoms at the beginning of the illness delays diagnosis [2].

Women older than 40 are encouraged by the National Breast Cancer Foundation (NBCF) to schedule mammography every twelve months. According to predictions, more than 2.2 lakh women in India were projected to receive the news of confirmation of breast cancer in 2023, and more than eighty thousand fatalities were anticipated [3]. The 2020 National Cancer Registry Programme Report estimates that by 2025, there will be more than 2.9 lakh new cases [4].

Consequently, according to the WHO, some behaviors and interventions such as continued infant feeding, consistent physical activity, keeping a healthy weight, avoiding drinking too much alcohol, and avoiding exposure to tobacco smoke could potentially reduce the chance of developing cancer.

Mammography, which involves taking a digital image of the breast, is employed to identify breast cancer. Since the technique necessitates a medical strategy which discovers tumors of the breast in female patients without causing any negative consequences, it is safe. The percentage of women who survive is higher for those who regularly undergo mammograms [5].

Over six million individuals died in 2018 as a result of breast carcinoma. Breast carcinoma excision screening and therapy including mammograms, ultrasounds, and biopsies consume quite a bit of time, hence a computerized diagnosis system utilizing a machine learning approach used [6, 7]. Machine learning algorithms and techniques help in early detection and screening. Unsupervised learning and Supervised learning are the two main categories into which the learning process in ML approaches may be differentiated. In supervised learning, a set of data instances that have been marked with the correct output are used to train the model [8-10]. Because there are no prepared data sets or expertise regarding the intended outcome in the situation of unsupervised education the objective is more challenging to accomplish. Unsupervised machine learning itself draws insights and hidden patterns from data [11]. One of the technologies that is frequently employed in supervised learning is ‘Classification’ [12]. It develops a model that uses information labeled to generate forecasts about the future [13].

In Figure 1, the survival rate has been shown based on a survey [14, 15]. Clinics and institutions in the medical industry keep substantial databases with patient medical histories, symptoms, and diagnoses [16]. The researchers use this information to design classification models by refining them with algorithms to generate predictions according to patient’s prior medical records [17]. Therefore, using the enormous amount of medical data that currently exists, medical inference has become much simpler with machine-based support [18].

All the methods used in this study fall under the category of binary classification models which diagnose whether the tumor is present or not based on certain features [19].

Patients with breast cancer have grown during the past ten years. Breast cancer may be the most common malignancy in women, with almost 2.3 million cases reported annually [20]. Breast carcinoma is the primary or secondary reason for death worldwide among women.

1.png

Figure 1. Breast cancer survivors' mortality rate (age-wise)

2.png

Figure 2. Breast cancer patients’ data (year-wise)

Female breast carcinoma has now passed carcinoma of the lungs as the illness that is most frequently diagnosed cancer [21]. There were 2,290,619 new cases of carcinoma of the breast globally during the year 2020 [22] and the graph getting wider every single day. The rise in the number of cancerous breast cancer patients from 2015 to 2023 is depicted in Figure 2 [23].

The above Figure 2 depicts that as per increasing year of age, the proportion of breast carcinoma patient’s also increased year-wise [24]. Survival rate was lowest for people >=75 years of age. In every year, increase in age, the chance of survival decreases, and the younger the age, the higher the chances of survival exist [25, 26].

In this section, proposed methodology is working with ensemble model with bagging on multiple machine learning models for analyzing the breast cancer patients in early phases so that the lifespan of these patients can be increased.

3.1 Dataset

The dataset for this breast cancer research has been taken from WBCD (Wisconsin breast cancer dataset), which has 569 instances with 30 attributes, where 357 are benign and 212 patients are malignant. This dataset was included with missing values in it and some of the data was unstructured.

Every parameter value in the WBCD data has a distinct numerical measurement value. With the exception of ID and diagnosis results, all other numerical measurement values for this investigation were normalized using the following calculation:

$\mathrm{Y}^{\prime}=\mathrm{Y}-\min (\mathrm{Y}) / \max (\mathrm{Y})-\min (\mathrm{Y})$           (1)

Here Y is an actual value and Y’ is the normalized value.

3.2 Performance evaluation metrics

In this research, ensemble model is used with bagging so after preprocessing of the dataset, following performance evaluation metrics [42, 43] is calculated:

$\begin{aligned} & \text { Accuracy }=(\text {True Positive}+ \text {True Negative}) / (\text { True Positive }+ \text { True Negative } + \text { False Positive}+ \text { Flase Negative})\end{aligned}$         (2)

$\begin{gathered}\textit {Precision}=\textit {True Positives} /(\textit {True Positives} + \textit {False Positives})\end{gathered}$          (3)

$\begin{array}{r}\textit {Recall}=\textit {True Positive} /(\textit {True Positive } + \textit {Flase Negative})\end{array}$          (4)

$\begin{gathered} \textit {F 1 Score}=2 *((\textit {precision} * \textit {Recall}) /(\textit {Precision} + \textit {Recall}))\end{gathered}$         (5)

Here in the above equations, True Positive, False Negative, True Negative and False Positive, are indicated for correct prediction and false prediction.

3.3 Bagging

This procedure, which is also known as an ensemble method, is bootstrap aggregating. Several models are trained individually on arbitrary portions of the data, and their predictions are then combined by averaging the outcome.

Following this, the bagging classifier trains several machine learning models separately on arbitrary data subsets before considering the median of their forecasts.

3.4 Model description

Machine learning models act as weak learners when applied individually on the dataset and the ensemble model works in a better way once applied on a dataset with machine learning algorithms so here in this proposed work, HEMBAGG is applied to machine learning algorithms like on KNN, decision tree, logistic regression, Random Forest, XG Boost, support vector machine etc to enhance the result.

Firstly, data extraction is performed, then data preparation, data analysis, implementation of the model, and the proposed hybrid model is defined. The Wisconsin Diagnosis Breast Cancer data has been employed in the present research to address the issue of prompt patient detection of breast cancer. The intention is to categorize and detect malignant individuals at the earliest stages. Implementing the suggested HEMBAGG entails the following measures.

In this research, a breast cancer dataset has been taken of 569 patients with 30 attributes from WBCD as defined in the methodology. Some of the machine learning algorithms like DT, KNN, LR, SVM, RF, XGB, etc. applied individually to this dataset. By using these algorithms, with the help of confusion matrix, it has been calculated that these algorithms are giving better accuracy than other machine learning algorithms as shown in Table 2 but by enhancing the algorithm with an ensemble model with bagging, it is improved.

According to the above described and proposed hybrid ensemble model, to forecast the labels for the test data (X_test), the model must first be trained through the initial training data (X_train and y_train) using an ensemble model with bagging. It is computed using the anticipated labels, and its results are compared to the actual labels (y_test). In the below figures, the confusion matrix score is displayed using an ensemble model with bagging on the decision tree for the WBCD dataset.

Table 2. ML algorithms accuracy on WBCD dataset

Figure 4 shows the metric scores for the ensemble model with KNN. Figure 4 shows the ensemble model's performance scores with KNN serving as the base estimator. Each score is shown as a labeled bar with the height of the bar corresponding to the score value. The graph makes it simple to evaluate and analyze the result of the ensemble model with the performance measures visually.

4.png

Figure 4. Metrics score for ensemble model using bagging with KNN

The results indicate that the overall derived precision rating is only 84.62% and accuracy 75.52% when the ensemble model with KNN is used to assess the accuracy, precision, recall, f-score, and support of patients classified as benign and malignant.

Figure 5 shows the ensemble model's performance scores with the Decision tree serving as the base estimator. Each score is shown as a labeled bar with the height of the bar corresponding to the score value. The graph makes it simple to finalize and analyze the result of the ensemble model with the performance measures visually.

5.png

Figure 5. Metrics score for ensemble model with decision tree

The results indicate that the overall derived precision rating is 98.11% and accuracy is 98.60%, when the ensemble model with the Decision tree is used to assess the accuracy, precision, recall, f-score, and support of patients classified as benign and malignant.

Figure 6 shows the ensemble model's performance scores with Random Forest serving as the base estimator. Each score is shown as a labeled bar with the height of the bar corresponding to the score value. This graph shows the score types for the ensemble model with Random Forest.

The results indicate that the overall derived precision rating is 91.23% and accuracy is 95.80%, when the ensemble model with Random Forest is used to assess the accuracy, precision, recall, f-score, and support of patients classified as benign and malignant. Using an ensemble model with a Random Forest gives an accuracy of 95.80%.

6.png

Figure 6. Metrics score for ensemble model with Random Forest

7.png

Figure 7. Metrics score for ensemble model with XGB boost

Figure 7 shows the ensemble model's performance scores with XGB boost serving as the base estimator. Each score is shown as a labelled bar with the height of the bar corresponding to the score value. The results indicate the accuracy i.e. 95.10% and the overall derived precision rating is only 91.07%, when the ensemble model with XGB boost is used to assess the accuracy, precision, recall, f-score, and support of patients classified as benign and malignant.

To increase the prediction score percentage for breast cancer patients, in the same algorithms, a HEMBAGG is used. Based on the given model, we can conclude that the decision tree is performing more efficiently and giving higher accuracy as compared to the rest of the algorithms shown in Table 3 and Figure 8.

Table 3. HEMBAGG (Ensemble model with bagging) accuracy score

8.png

Figure 8. Accuracy comparison of ml algorithms vs. HEMBAGG model

Table 4. Comparative accuracy score with ML algorithms and HEMBAGG model

In above Table 4, the accuracy score of ML algorithms is compared with the HEMBAGG model, and comparatively high accuracy is obtained with the new hybrid model (HEMBAGG) that is shown in Table 4, and the graph is also displayed below with the accuracy score of both of the algorithms.

HEMBAGG model overcomes the drawbacks of the poor generalization of the traditional machine learning models by overcoming the problem of over-fitting and providing better accuracy, robustness, and generalization.

Table 4 demonstrates the accuracy comparison of ML Algorithms like DT, KNN, LR, SVM, RF, and XG Boost with the Accuracy of the HEMBAGG model. In most of the algorithm cases the accuracy is not being changed but and Decision Tree and K-nearest neighbor it is changed and the HEMBAGG model with the Decision Tree gives the highest prediction.

Table 5. Accuracy score of decision tree with different parameters

The above Table 5 represents the accuracy score of the decision tree which is a base estimator with different parameters. So, for this confusion matrix is calculated first and it is found that the ensemble method with bagging on decision tree classifier is giving a very high accuracy.

Figure 9 shows the graph of labels of accuracy scores of the Decision tree with different parameters. Each score is shown as a labeled bar, with the height of the bar corresponding to the score value. The graph provides a visual comparison of the performance metrics, allowing for easy interpretation and analysis of the ensemble model's effectiveness with the proposed ensemble method with bagging which enhances the result.

9.png

Figure 9. Accuracy with HEMBAGG with decision tree

The actualization of the code for the hybrid ensemble model with a bagging classifier is employed to forecast breast carcinoma patients which uses a development framework with Python, also known as Jupyter with Pandas library. This environment has advanced in the numerical environment. The ensemble model is an enhanced version of the decision tree and with bagging, it determines the accuracy, precision, recall, and f1-score. The result shows that the overall derived precision rating of classifiers is 92.31% and 98.6%, respectively, as shown in Table 5 and Figure 9. So, we can conclude that the HEMBAGG model has increased the accuracy using the ensemble method so that people's lives could be saved in the early stage of cancer.

In the end, it is concluded that in various research independent machine learning algorithms have been applied but, in those researches, the result is not as efficient as it is predicted with the proposed model i.e., ensemble model with bagging because the ensemble model is a group of multiple algorithms. It generates better results compared with already existing research.