CANCER CELL CLASSIFICATION USING MACHINE LEARNING

CHAPTER ONE: INTRODUCTION
1.1 Background of the Study
Cancer cell remains one of the leading causes of morbidity and mortality worldwide, with millions of new cases diagnosed each year (World Health Organization, 2024). It encompasses a wide variety of diseases characterized by the uncontrolled growth and spread of abnormal cells, which can invade nearby tissues and metastasize to distant body sites.
The early detection and accurate classification of cancer cells are critical for effective treatment and improved patient survival rates (Jiang et al., 2023).
A collection of disorders is called cancer. It entails unnatural cell development.
It spreads quickly and has an impact on other bodily parts.
Not every type of tumor is necessarily malignant.
Some tumors do not spread throughout the body. Numerous symptoms of cancer include tumors, unusual bleeding, persistent coughing, increased weight loss, etc. Nearly 100 different varieties of cancer can afflict a person.With the advancement of Artificial Intelligence, particularly Machine Learning (ML), cancer diagnosis can now be automated with higher accuracy.
Convolutional Neural Networks (CNNs) have emerged as powerful tools for image-based cancer classification, capable of learning complex patterns directly from medical images.
Cancer research is one of the difficult, alluring, and important areas of attention in the medical industry.
Accurate automatic tumor and cancer prediction systems are required in order to give patients the proper care.
The earlier therapies were manual and clinically based.
The following are some drawbacks of these traditional categorization methods: - Slow diagnosis procedure, cancer prognosis based on pathology reports, which may raise the patient's efficacy Different kinds of clinical courses are needed for cancer classification and prediction.
Various machine learning methods are utilized in the contemporary world to address various issues and design automated systems.
These technologies are able to identify and categorize cancer in photos or patient information like age Traditional diagnostic methods, such as histopathological examination by pathologists, although effective, are labor-intensive, time-consuming, and prone to human error due to the subjective nature of manual assessment (Shen et al., 2022).
Skin cancer arises from the uncontrolled growth of abnormal skin cells, often triggered by unrepaired DNA damage leading to mutations.
The population at a higher risk of developing this chronic illness includes those with white skin due to their reduced melanin levels [5].
It is considered crucial for the healthcare systems of various countries to allocate resources for skin cancer prevention. For instance, Australia has implemented mass media campaigns accompanied by policies and access to products designed to protect people from ultraviolet radiation, adhering to national standards
The rapid growth of medical imaging technologies and biomedical data acquisition has provided vast amounts of data, which, when coupled with advanced computational techniques, have opened new opportunities for automation in cancer diagnosis.
Machine Learning (ML) has emerged as a transformative approach to medical diagnostics, particularly in cancer cell classification.
By training algorithms on large datasets of labeled medical images or cell data, ML models can learn complex patterns that distinguish between malignant and benign cells, often with accuracy comparable to or exceeding human experts (Kourou et al., 2023).
Traditional ML methods, such as Support Vector Machines (SVM), Decision Trees, and Random Forests, have shown success in structured datasets such as the Wisconsin Breast Cancer Dataset.
In recent years, Deep Learning (DL) techniques, particularly Convolutional Neural Networks (CNNs), have revolutionized image-based cancer detection by automatically extracting hierarchical features from raw images (Litjens et al., 2017).
The need for robust, accurate, and real-time cancer detection systems is increasingly critical as healthcare systems face rising workloads and demand for faster diagnostics.
This study focuses on developing and comparing different ML models to classify cancer cells, with the aim of identifying the best-performing approach for reliable, efficient, and scalable diagnostic support. and symptoms.
The field of cancer research is not new to the use of machine learning. Since the last 20 years, decision trees and artificial neural networks (ANNs) have been widely used in various industries for early cancer detection and diagnosis. One of the recent PubMed surveys offers data on the amount of research done on cancer detection techniques.
It demonstrates that papers based on the idea of the relationship between cancer and machine learning are published in numbers of at least 1500.
Many studies focus on using machine learning techniques for identifying, classifying, and detecting cancer or tumors. Machine learning was once exclusively beneficial for detecting and diagnosing cancer, but more recent algorithms have mostly concentrated on cancer prognosis and prediction. The most popular machine learning techniques are data mining algorithms, which are used to classify gene expression data and cancer. The purpose of the scenario and cancer prophecy is to: 1) The first step in predicting cancer susceptibility is to determine the likelihood that a certain type of cancer will develop before it manifests itself. 2) After giving a remedy for the original cancer, the extrapolation of cancer recurrence determines the likelihood of cancer development again. 3) After a cancer diagnosis, the likelihood of cancer survival is determined by looking at factors including life expectancy, survival, progression, tumour drug sensitivity, etc. In the modern internet age, websites for healthcare or social media create a lot of data.
Data mining techniques can be used to extract symptoms from this vast amount of data, which will be further helpful for cancer detection or classification.
The two most crucial methods utilised for symptom analysis and cancer classification are clustering and classification.
This essay provides an overview of recent studies that use both online and offline data to classify cancers. This survey covers the most recent findings in the cancer prognosis and prediction. In this paper, Section II will discuss the literature review, Section III will describe the proposed system, and Section IV will provide a conclusion.
Such evidence highlights why melanoma remains a significant public health concern globally and emphasizes the importance of promoting and supporting research dedicated to obtaining and developing techniques that contribute to timely detection and accurate diagnosis.
Delayed diagnosis can allow cancer to progress unchecked, leading to more advanced disease stages, increased morbidity, and reduced treatment options
1.2 Statement of the Problem
Despite advances in cancer research and diagnostic technology, several challenges persist in cancer cell classification:
1. Diagnostic errors and delays – Manual examination of cancer cell samples is prone to human error and often limited by the availability of skilled pathologists.
2. Variability in sample quality – Image quality, staining techniques, and patient-specific differences can cause inconsistency in diagnosis.
3. Scalability limitations – Traditional manual diagnostic methods cannot efficiently handle the growing demand for rapid, large-scale screening.
4. Model generalization issues – Existing automated systems sometimes fail to generalize well to new, unseen data due to variations in cell morphology across populations and imaging devices.
Machine learning offers promising solutions to these problems by enabling automated, consistent, and high-accuracy cancer detection. However, there is a lack of comparative studies evaluating the performance of different algorithms across diverse cancer cell datasets, particularly those using both traditional and deep learning methods.
1.3 Aim and Objectives of the Study
• To develop and compare the performance of different machine learning algorithms in classifying cancer cells, identifying the most accurate and efficient approach for practical medical use.
Objectives
• To preprocess and prepare cancer cell datasets for machine learning classification.
• To develop and train multiple ML models (e.g., CNN, SVM, Decision Trees) for cancer cell classification.
• To evaluate the models using performance metrics such as accuracy, precision, recall, and F1-score.
• To identify the best-performing model for potential real-world deployment in clinical decision support systems.
1.4 Scope and Limitation of the Study
The study focuses on classifying cancer cells as benign or malignant using publicly available datasets, such as histopathological images and structured datasets (e.g., Wisconsin Breast Cancer Dataset). The scope includes data preprocessing, model training, evaluation, and comparative analysis.
The primary goal of many artificial intelligence (AI), machine learning, and data science is the
discovery of new facts from data based on statistical and logical methods. The secondary goal of
these disciplines is to communicate the new facts [Aumann et al., 2003][Dhar, 2013]. Of course,
the discovery should be valid and reproducible. Unfortunately,many reported discoveries are not
reproducible due to sloppy data preparation and clean up [Editors, 2012][Economist, 2013].
Typically, many projects use data sets that were not necessarily collected for those projects. For
example, SEER database is built for summarizing cancer data and not survivability prediction. The survivability prediction problem is a binary classification with uneven distribution of data points [Vapnik, 1995][Xiao et al., 2009]. In order to prepare SEER data for binary classification, we must first decide how to assign data points to each class. According to Parkin and Hakulinen, a well accepted methodology in predicting patient survival involves summarizing and analysis [Parkin and Hakulinen, 1991]. The most widely used metric involves calculating the percentage of patients alive after five years, using a direct method as outlined by Parkin and Hakulinen [Parkin and Hakulinen,1991]. Chapter 3 gives our detailed explanation of our approach to data assignments based on direct method.

Limitations include:
Use of publicly available datasets that may not cover all cancer types or global demographic variations. High computational requirements for training deep learning models. Focus on binary classification rather than multi-class cancer type identification.
Justification of the Study
Early and accurate detection of cancer significantly improves patient survival rates and reduces healthcare costs. Automated classification systems powered by machine learning can serve as decision-support tools for pathologists, reducing diagnostic time and minimizing human error. This study’s comparative approach will provide valuable insights into selecting the most effective ML method for cancer cell classification, contributing to the advancement of AI-assisted healthcare diagnostics.

Definition of Terms
Cancer Cell – An abnormal cell that grows uncontrollably and may invade nearby tissues or spread to other body parts.
Machine Learning (ML) – A branch of artificial intelligence where systems learn from data to make predictions or decisions without explicit programming.
Convolutional Neural Network (CNN) – A deep learning architecture widely used in image analysis tasks, including medical image classification.
Support Vector Machine (SVM) – A supervised machine learning algorithm used for classification tasks that finds an optimal boundary between classes.
Precision – The proportion of correctly predicted positive cases among all predicted positive cases.
Recall (Sensitivity) – The proportion of correctly identified positive cases among all actual positive cases.


CHAPTER TWO: LITERATURE REVIEW
2.1 Traditional Machine Learning Approaches in Cancer Cell Classification
Epimack Michael et al. proposed a framework for classifying cancer cell using machine learning that has been optimized for improved performance.
This paper takes a different stance in setting the values of the hyperparameters.
They propose an automatic ML that can automatically build optimized ML algorithms rather than using default values.
Traditional machine learning techniques have been applied extensively in cancer diagnosis, particularly in structured datasets such as the Wisconsin Breast Cancer Dataset (WBCD).
These methods rely on handcrafted features derived from clinical or histopathological data, such as cell size, texture, nucleus shape, and chromatin density.
Common algorithms include Support Vector Machines (SVM), Decision Trees (DT), Random Forests (RF), k-Nearest Neighbors (k-NN), and Logistic Regression (LR).
For example, SVM has been successfully applied to WBCD, achieving accuracy above 96% in some studies (Patil & Kumar, 2022). Random Forests have demonstrated strong generalization by averaging predictions over multiple decision trees, reducing overfitting and improving robustness (Kourou et al., 2023). Logistic Regression has also been used for quick and interpretable cancer detection models, though it tends to underperform with non-linear data.
This study [4] develops a new deep learning (DL) model based on the transfer learning (TL) technique to effectively support the automatic detection and diagnosis of the BC suspicious region based on two techniques, namely 80-20 and cross-validation.
DL architectures are designed to focus on particular problems.
TL applies the knowledge acquired while resolving one issue to another pertinent issue.
The proposed model for mammographic image analysis uses pre-trained convolutional neural network (CNN) architectures, including Inception V3, ResNet50, Visual Geometry Group networks (VGG)-19, VGG-16, and Inception-V2 ResNet, to extract features from the MIAS dataset. By classifying mammogram breast images Experimental results show the effectiveness of the proposed model, with overall accuracy, sensitivity, specificity, precision, F-score, and AUC of 98.96%, 97.83%, 99.13%, 97.35%, 97.66%, and 0.995, respectively, when using the 80-20 method, and 98.87%, 97.27%, 98.2%, 98.84%, 98.04%, and 0.993 when using the 10-fold crossvalidation method.
A computer-aided diagnostic (CAD) [5] approach for classifying patients into three categories (cancer, no cancer, and non-cancerous) under the control of a database has been proposed in this study.
The classification stage has been analysed using Convolution Networks, Support Vector Machines (SVM), and Random Forest, three efficient classifiers (RF).
The segmented thermographic pictures have been subjected to analysis and classification using a Convolutional Neural Network (CNN).
This method divides the image input variables into pre-expanded RGB and Gray channels, which are then combined with an autonomous image denoising and classification.
The resulting output will feed both processes to the breast image analysis and feature extraction network, where a two-extraction determines whether the image is benign or malignant.
The accuracy that CNN attained was 99.67 percent, compared to the accuracy that SVM and RF attained, which were both 89.84 percent and 90.55 percent, respectively.
This paper [6], presents a deep learning-based technique for the segmentation and detection of colorectal cancer from digitized H&E-stained histology slides.

    Date: 2026-08-02 00:00:00.000000