Efficient Breast Cancer Dataset Analysis Based on Adaptive Classifiers

1. INTRODUCTION

The development of the Internet and its containment of a vast amount of data, coupled with the emergence of large data volumes generated by social media, has made data today exceptionally massive. This necessitates new approaches for its transfer, processing, and analysis, collectively known as big data. This data may be unstructured, semi-structured, or structured, benefiting various sectors such as societal, business, industry, agriculture, education, and healthcare [1], [2].

The internet of things (IoT) contains a huge number of sensors connecting to other systems and devices, producing extensive data that requires novel analytical methods, particularly wireless sensor networks, as the data source [3], [4].

Given the immense volume of data, traditional database systems are inadequate for storing, handling, and analyzing such quantities. The concept of big data involves large-scale processes for identifying and interpreting information into new insights. While the term “big data” has long been in use, its prominence surged following the rise of social media [5], [6]. The exponential growth of data worldwide, characterized by vast volume, high speed, and diverse types, demands an infrastructure capable of simultaneous processing and storage. Cloud computing offers on-demand computing resources that can be quickly configured, provisioned, and released as per users’ needs, making it accessible to individuals and organizations alike [7], [8]

In data analysis, starting from processing, cleaning, and filtering, correct data retrieval is essential to perform the necessary analysis. Analyzing and calculating metrics such as maximum, minimum, and standard deviation using relevant tools are crucial for obtaining accurate results. Optimal analysis follows proper steps to ensure accuracy and reliability [9], [10]. Smart cities can utilize distributed stream processing frameworks for real-time data processing, a practical application in addition to their current IoT adoption. Choosing a suitable framework for smart city data analytics requires a comprehensive understanding of target application features [11], [12].

Big data analysis employs various analytical methods depending on the data’s nature, aiming to extract insights. These methods include statistical analysis, correlation analysis, and regression analysis [13], [14]. The aim of the proposed approach is to develop an effective algorithm for disease prediction by employing classification approaches to address health-care issues during diagnosis and achieve efficient performance results.

2. LITERATURE REVIEW

Babar et al. proposed a health-care architecture based on energy harvesting analysis of health monitoring devices. The proposed architecture consists of three layers. Consistent datasets were verified on Hadoop server to validate the proposed design based on the threshold value. The goal of this research is to make smart decisions and deal with events. The analysis shows that the proposed design has great potential in the field of smart health [15].

Singh and Yassine utilized IoT and big data analytics to create energy management strategies to manage home energy effectively and efficiently. They proposed a unified architecture that enables creative activities to process massive amounts of granular energy use data in close to real-time. The complexity and resource requirements of data processing, storage, and classification analysis in close to real-time are addressed by proposing an IoT big data analysis system that uses fog computing [16].

Shah et al. proposed a new design and philosophy for the disaster-resistant smart city. Together, the Hadoop ecosystem and Spark form a powerful system environment that allows for real-time and offline analysis. System efficiency is evaluated in terms of processing time and throughput. The aim of this research is to add to the body of knowledge and guide future research on the design and implementation of disaster-resilient smart cities based on this system and the IoT. This strategy can lead to immediate and effective situational awareness, which can help mitigate the effects of the disaster [17].

Syed et al. proposed a new smart healthcare framework to monitor the physical activity of older individuals using the Internet of Medical Things and intelligent machine learning algorithms for quick analysis, decision-making and better treatment suggestions. Hadoop MapReduce algorithms are used to handle massive amounts of data in parallel. The aim of this study is to predict the physical activity of respondents to help them live a healthier lifestyle. This study provides an excellent option for detecting physical activity and monitoring the health of elderly people remotely [18].

Li introduced a fog-assisted IoT-based intelligent and real-time health-care information processing system. In this system, minimal latency and large amounts of data generated by IoT sensors are offloaded to the fog cloud for data analysis and processing. The data then are processed and stored in a central cloud system using Hadoop and Apache Spark in order to process and analyze. The proposed compression strategy results in a 60% reduction in the amount of data in this system, in addition, for real-time data analytics; it offers a fog-powered approach with a big data environment [19].

Hadi et al. presented a multidisciplinary approach to e-health care, priority, big data analytics, and radio resource optimization in a multi-tier 5G network. They used a combined system including three machine learning algorithms (naïve Bayesian classifier, logistic regression, and decision tree) to evaluate historical outpatient stroke medical data and signals from IoT sensors attached to the body to predict the probability of an impending stroke. Two optimization approaches, namely, the weighted sum rate maximization approach and the proportional fairness approach, are presented to achieve this goal. The proposed methods improved the average signal-to-noise ratio and proportional fairness [20].

Ge et al. demonstrated a system that collects data from sensors and used deep learning to evaluate and monitor patients’ health data to predict disease and provide timely alerts in this work. Extensive analysis and experimental data are provided to show that the proposed strategy was safe and effective. When a patient uploads their health information, the system allows for precise access and confirmed deletion [21].

Lv et al. developed the theory of fuzzy function-based mean clustering algorithm using K-means and fuzzy theory in big data analysis technology. The results indicated that when the effective propagation probability was 100% and the value was between 0.01 and 0.05, it was close to the true result and has the least data delay. This study showed that using big data analysis techniques to enhance electric vehicle transportation networks can significantly reduce network data transmission performance delay and adjust the route to effectively stop the spread of congestion [22].

Kaya et al. studied priority of speed and accuracy in health data and aimed to detect anomalies at the edge of IoT for effective management of big data. The data stream for age, gender, height, time, temperature, and weight is used in the analysis through applied Naïve Bayes, neural network, logistic regression, and random forest algorithms. The experimental results compared speed and accuracy and obtaining logistic regression (LR) algorithm provides great success in the IoT system. Machine learning (ML) algorithms are suitable for the IoT edge because they can make timely and efficient decisions in the health-care sector [23].

Ahmed et al. designed an approach based on neural networks, which aimed to diagnose and predict the epidemic. This model provided descriptive, diagnostic, predictive, and prescriptive analysis using big data analytics. This model was used to predict COVID-19 within the creation of a health monitoring platform. The results of the neural network-based model were also compared with the results of other machine learning methods. The neural network-based algorithm achieves high percentage accuracy without using any computationally expensive deep learning-based methodology [24].

Qin et al. used big data analytics to implement the multimedia-assisted student-centered learning model, which addresses critical policies to help educational institutions achieve this transformation in a more systematic way by clarifying teachers’ instruction. When compared to alternative technologies, simulation results indicate that the proposed model improves student attention (97.3%), efficiency (90.1%), student retention rate (97.5%), engagement (98.2%), and learning outcomes (95.3%) [25].

3. DATA SET

The University, of Wisconsin, (USA) provided this breast cancer database. This dataset consists of 699, instance and 10, attributes, the target (class): (two for benign and four for malignant). The dataset contains (458) benign, class and (241) malignant class so that the data of dataset are imbalanced data. The description attributes with its values about this dataset can be summarized as follows:

Sample code number, clump thickness (1–10), uniformity, of cell size (1–10), uniformity of cell shape (1–10), marginal adhesion (1–10), single, epithelial cell size (1–10), bare nuclei (1–10), bland chromatin (1–10), normal chromatin (1–10), mitoses (1–10), and class (two for benign and four for malignant).

4. PROPOSED APPROACH

The dataset including information about the situation of patients, then will be processed using prediction approach such as Kernal Naive Bayes, Linear support vector machine (SVM), Coarse Tree, K-Nearest Neighbors, Cosine KNN, Coarse Gaussian SVMs and Fine Tree. Then, the performance of the algorithms will be evaluated using a suitable evaluation model such as k-fold cross-validation and percentage split approaches. The proposed approach includes many parts as shown in Fig. 1.

Fig. 1. Proposed approach for breast cancer dataset.

5. DISCUSSION AND ANALYSIS

The of breast cancer confusion matrix is implemented in the case of percentage split which is used to calculate different performance metrics such as precision, Recall and f1-score as shown in Fig. 2.

Fig. 2. Confusion matrix of breast cancer (percentage split).

The precision, recall and f1-score are measured for Kernel Naïve Bayes, Linear SVM, coarse tree classifiers respectively. Fig. 3 shows the performance comparison values of Kernel Naïve Bayes, Linear SVM, and Coarse Tree classifier algorithms in case of percentage split.

Fig. 3. Comparison of kernel NB, linear support vector machine, coarse tree algorithms.

In this work, different classifier algorithms are applied on breast cancer dataset. The Kernel Naïve Bayes is adapted to classify the breast cancer and the outcomes of the experimented results are compared the precision, recall and f1-score of many classifier algorithms like Coarse Tree and Linear SVM. So that the obtained results are 100% for all precision, recall and f1- score for Kernel Naïve Bayes approach.

Applying different classification algorithms on breast cancer dataset leading that the subspace KNN is adapted to classify the breast cancer in case of 4-fold cross validation. Fig. 4 shows the confusion matrix which used to compute the different performance metrics such as accuracy, sensitivity, specificity and f-measure.

Fig. 4. Confusion matrix of breast cancer (4-fold cross validation).

Precision, recall, and f1-score are measured for Subspace KNN, Cosine KNN, Coarse Gaussian SVM and Fine Tree classifiers. Fig. 5 shows the performance comparison values of Subspace KNN, Cosine KNN, Coarse Gaussian SVM and Fine Tree classifier algorithms in case of 4-fold cross validation. This figure indicated a better performance in case of subspace KNN compared to other algorithms.

Fig. 5. Metrics of breast cancer (4-fold cross validation).

Table 1 shows the outcomes of the proposed approach compared with other studies that used this dataset. The classification algorithms used in this study show that the obtained performance is 100% for precision, recall, and f1-score compared to reference [7]. In addition, the performance is 97.2%, 99.1%, and 98.1% for precision, recall, and f1-score compared to reference [8].

TABLE 1: Results are compared to previous results in breast cancer dataset

6. CONCLUSIONS

Efficient big data analytical approaches and machine learning algorithms are utilized for accurate prediction and decision making. So that different machine learning algorithms are being implemented for predicting breast cancer. The machine learning techniques are using comparisons based on important performance metrics such as precision, recall, and f1-score. Machine learning (disease prediction) is a suitable technique that assists in the early detection of disease, may aid practitioners in diagnosis decision-making, and also gives accurate predictions. Many machine learning algorithms applied on Wisconsin breast cancer (WBC) dataset that generate best performance accuracy for diagnosis and prediction the WBC dataset, in which the Kernel Naïve Bayes gives 100% as precision in percentage split method, the Coarse Gaussian SVM is adapted to classify the breast cancer in case of 4-fold cross validation which gives 97.2% as precision.