Table 1.

Review of various cancer prediction techniques.

Sr. no.	Paper name	Objective	Technique/tool	Dataset	Findings
1	[8]	To design a method to classify and predict classes of cancer	Neighborhood analysis, DNA microarrays, self organizing maps	27 ALL samples from Dana-Farber Cancer Institute, 11 adult AML samples from the Cancer and Leukemia Group B (CALGB) leukemia cell bank	Feasible method. Proper experimental care is required
2	[2]	To classify samples of cancer for gene expression data	Computational analysis, affymetrix oligonucleaotide arrays, neighborhood analysis, genecluster software	38 leukemia samples (11 AML, 27 ALL), for testing 34 samples (14 AML, 20 ALL)	Genes with no correlation provide a better result, and the median prediction strength is 0.86
3	[9]	To specify the specific categories of cancer using their gene expression	ANNs, cDNA microarrays, DeArray software	NCI, ATCC, MSKCC, CHTN, DZNSG, National Institutes of Health	It can work with nonlinear features also. It is robust. It also achieves high sensitivity and specificity.
4	[10]	To create a framework for predicting predefined classes of tumor	Compound covariate prediction, BRB ArrayTools	Hereditary breast cancer dataset of 22 patients [11]	Good setter for comparing prediction methods. Require some improvements.
5	[12]	To develop a classification system for DNA microarray gene expression data	SOMs, Cluster and TreeView software, PCA, KNN	Multiple datasets have been used, such as one with 99 samples, the other with 42 selections,	Gene expressions provide an excellent way of diagnosing patients with medulloblastomas
6	[13]	To propose a method that performs classification on interval-scaled attributes basis	PCA, FA, fuzzy FA	203 samples (a subset of the actual dataset used in [14])	Successfully used in supervised learning. FA provides more information compared to surgical-pathological staging
7	[15]	To propose a method for gene feature selection	Multiple SVM-RFE	Four gene expression datasets available on Kent Ridge Bio-Medical Data Set Repository	MSVM-RFE has classification accuracy better than SVM-RFE. SVM's performance has been improved.
8	[16]	To propose a framework for addressing the problem of integration of different data types	Generalized singular value decomposition	Fourteen breast cancer cell lines from American Type Culture Collection	Gene expression and copy number data are being analyzed. Improvements can be made to use other data types also.
9	[17]	To propose a method used to find tissues of the tumor with different gene expression data	ssEAM, PSO	NC160, acute leukemia, ALL dataset	ssEAM performs better than PNN, ANN, LVQ1and KNN at a 0.05 significance level
10	[18]	To present a selection method for analyzing gene expression data	RBF neural network, rough based feature selection method, naïve Bayes, linear SVM	ALL, AML, lung cancer and prostate cancer dataset (http://sdmc.lit.org.sg/GEDatasets/Datasets.)	The best classification accuracy rate of 99.8%
11	[19]	To present a framework for discovering cancer classes.	Permutation technique, cluster ensemble, cluster validity index (DAI)	3 synthetic and 4 real datasets (leukemia [2], Novartis multitissue [20], lung cancer [14], St. Jude [21])	DAI finds the number of classes correctly and outperforms other existing methods
12	[22]	To present a method based on gene expression for classifying NSCLC	Hierarchical clustering, SpotFire decision site, proportional hazards model	91 NSCLC, six normal lung tissues from GSE3526 (Duke University)	Gene signatures provide the best way for histopathological classification
13	[23]	To propose a classifier predicting disease in CRC patients	Agilent 44K oligonucleotide arrays, Kaplan–Meier method, unsupervised hierarchical clustering	188 training samples (NCI, LUMC, SGH) and 206 testing samples (Institute Catalad'Oncologia, Spain)	Eighty-six percent of patients of the validation dataset are identified as low-risk patients. First prognostic technique for CRC
14	[24]	To propose a framework that combines genome-wide copy number and expression data	L1-L2 constrained regression, local and global search strategies	89 samples of breast cancer Dataset (UG San Francisco and California Pacific Medical Center [25])	Outperforms other existing methods accuracy
15	[26]	To propose a framework that combines other models that describes gene interaction.	Bayesian model, Gibbs distribution, ANOVA test, parallel programming with GPU/CPU	GSE4290, DREAM dataset	Specificity of 0.99 has been achieved. Better performance than Enet and VAR
16	[27]	To propose the extended framework for segmentation of breast tumor	Multichannel MRFs, kinetic observation model, Gaussian mixture model	DCE MRI images of breast cancer	AOC of 0.9 has been achieved using multichannel MRF compared to AOC of 0.89 in single-channel MRF. Better segmentation results when applied to SVM
17	[28]	To propose a gene selection method	LSLS, wrapper method, SVM	Six datasets available at Kent Ridge Biomedical Data repository	LSLS performs better than KW and SPFS
18	[29]	To present a novel method classifying tumor samples.	RPCA, LDA, SVM	Nine different publically available datasets (acute leukemia data [2], colon cancer data, glimos data, medulloblastoma data, prostate cancer data, 11_tumor data, and brain tumor data)	Performance is measured using LOO-CV, accuracy, and AUC. A feasible and effective method.
19	[30]	To propose a method based on deep learning for inferring target genes expression	D-GEX	Microarray GEO dataset, RNA-Seq-based GTEx dataset	Outperforms linear regression (15.33 relative improvement) and KNN. The lower error rate in most of the genes (81.31%).
20	[31]	To develop a fused network identifying KIRC stages	Gene expression and DNA methylation data, SNF, SNFTool, sparse partial least square regression, LASSO label prediction method	The Cancer Genome Atlas KIRC data (TCGA data portal)	High prediction accuracy than KNN, MLW, and WDC. It is robust.
21	[32]	To classify widely and rarely expressed genes	Incremental feature selection method, mRMR, RNN	Gene expression dataset available at the Human Protein Atlas [33]	GO terms and KEGG are used at the functional level. Youden's indexes are 0.739 and 0.639 for normal and cancer tissues, respectively.
22	[34]	To develop a light-weight CNN for classifying breast cancer	CNN, array-array intensity correlation, R-Studio, batch normalization	Breast cancer dataset from Pan-Cancer Atlas	Achieves 98.76% accuracy
23	[35]	To propose a method for classifying different types of cancer.	BPSO-DT, CNN, deep learning	Cancer types: RNA sequencing values from tumor samples/tissues available at Mendeley datasets	It achieves an accuracy of 96.90%. Various evaluation parameters are recall, precision, and F1 score.
24	[36]	To propose a method based on NMF to classify tumor	NMF, SNMF, SVM	Colon cancer dataset [37], acute leukemia dataset, medulloblastoma dataset	It is effective and efficient. The effect of sparseness is low.
25	[38]	To propose a model for biclustering data of gene expression.	PCA, GLPCA, DHPCA,	SRBCT, medulloblastoma, colon cancer, 11_Tumors	It is compared with PCA, GLPCA, GNMF, ONMTF, and NMTFCoS. It provides better accuracy than others.
26	[39]	To present a framework for predicting the expression of genes employing nonlinear features	Unsupervised clustering algorithm, L-GEPM, LSTM neural network	GEO data from LINCS cloud, GTEx, and 1000G RNA-Seq data	Performs better than D-GM, LR-L1, and KNN-R. Target genes extracted are much closer to the actual gene expression. Flexible and superior for NL features.
27	[40]	To propose a multilayer framework to classify multitissues of cancer.	CNN, RNA sequencing, supervised learning, stochastic gradient descent optimization, back-propagation	11093 samples from the Cancer Genome Atlas	98.93 percent overall accuracy and 0.99 AUC have been achieved
28	[41]	To propose a gene selection method that can classify tissues in multicategory datasets	PLS, linear support vector classifier, MATLAB, OSU_SVM3.00 toolbox linear SVC, SVM	MIT AML and ALL dataset, SRBCT datasets	It is efficient and robust. It works well for both two-category and multicategory datasets.
29	[42]	To propose an ST model for finding the effects of CNAs	LST and NA, dynamic modeling, transcriptional bursting, transcriptional oscillation, circular binary segmentation	NCBI/GEO database	It shows the use of mathematical theory to investigate the findings and for a better understanding of cancer bio
30	[43]	To propose a muti-fusion-based method for profiling gene expression under nonthermal plasma treatment.	Dempster–Shafer method, fuzzy C-Means clustering method, MATLAB R2016b	NCBI Gene Expression Omnibus under GEO (GSE59997)	Reduces uncertainty and increases reliability. The use of C-means finds changes in genes in various nonthermal plasma treatments.
31	[44]	To present a survey of 1D CNN and its applications.	NA	NA	1D CNN works well with small data and where fewer computations are required. It also works where low-cost implementation is needed.
32	[45]	To propose a classification method for ECG signal images based on 2D CC.	CNN, Intel17-5930K CPU, and NVIDIA GTX1080 GPU	MIT-BIH Arrhythmia database	2D CNN outperforms 1D CNN. 2D CNN is more accurate and robust. 1D CNN works well with limited data.