Chapter 16 - An enhanced feature selection and cancer classification for microarray data using relaxed Lasso and support vector machine

Background Machine learning (ML) algorithm support vector machine (SVM) performed better than Thrombolysis in Myocardial Infarction (TIMI) score for ASIAN STEMI patients. However, Deep Learning (DL) effectiveness in the multiethnic ASIAN population has yet to be determined. DL has automatic learning of the feature from a given dataset without the need to conduct feature selection. However, the selected features by the algorithm is black box. Identifying features associated with mortality is essential to recognize characteristics of patients with high risk for better patient management. Purpose To develop a DL algorithm for in-hospital mortality in multiethnic STEMI patients using predictors identified from the SVM algorithm. To investigate DL performance constructed using predictors from SVM feature extraction and expert-recommended predictors. Methods We constructed four algorithms; a) DL and SVM algorithms with predictors identified from the SVM variable importance b) DL and SVM using predictors based on expert recommendation. We used registry data from the National Cardiovascular Disease Database of 11397 patient's. Fifty parameters including demographics, cardiovascular risk, medications and clinical variables were considered. The Area under the curve (AUC) is the performance evaluation metric. Algorithms were validated against the TIMI and tested using the same validation data. SVM variable importance with backward elimination was used to select and rank important variables. Results DL algorithms outperform SVM and TIMI on the validation dataset; i) DL with SVM selected predictors (15 predictors, AUC = 0.97), ii) DL with expert-recommended predictors (16 predictors, AUC = 0.96), iii) SVM with selected predictors (15 predictors, AUC = 0.92), iv) SVM with expert-recommended predictors (AUC = 0.89) and TIMI (AUC = 0.82). Common predictors across SVM feature selection, expert-recommendation and TIMI are: age, heart rate, Killip class, fasting blood glucose, systolic blood pressure, comorbid diseases and ST-elevation. SVM feature selection also identified diuretics, PCI and pharmacotherapy drugs as predictors that improve mortality prediction in STEMI patients. Our findings suggest that the TIMI score underestimates patients risk of mortality. DL algorithm using selected predictors classified 35% of nonsurvival patients as high risk (risk probabilities >50%) compared to only 12.7% nonsurvival patients by TIMI (score >5) (Figure below). Conclusions In the ASIAN population, patients with STEMI can be better classified using the DL algorithm compared to the ML and TIMI score. Combining ML feature selection with DL allows the identification of distinct factors in a unique ASIAN population for better mortality prediction than relying solely on an expert recommendation as it is a very subjective approach. Continuous validation on population-specific algorithms using DL and ML is needed before implementing in a real clinical setting. Funding Acknowledgement Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Technology Development Fund 1 TIMI performance on validation setDL performance on validation set

BackgroundMachine learning (ML) algorithm support vector machine (SVM) performed better than Thrombolysis in Myocardial Infarction (TIMI) score for ASIAN STEMI patients. However, Deep Learning (DL) effectiveness in the multiethnic ASIAN population has yet to be determined. DL has automatic learning of the feature from a given dataset without the need to conduct feature selection. However, the selected features by the algorithm is black box. Identifying features associated with mortality is essential to recognize characteristics of patients with high risk for better patient management.PurposeTo develop a DL algorithm for in-hospital mortality in multiethnic STEMI patients using predictors identified from the SVM algorithm. To investigate DL performance constructed using predictors from SVM feature extraction and expert-recommended predictors.MethodsWe constructed four algorithms; a) DL and SVM algorithms with predictors identified from the SVM variable importance b) DL and SVM using predictors based on expert recommendation. We used registry data from the National Cardiovascular Disease Database of 11397 patient's. Fifty parameters including demographics, cardiovascular risk, medications and clinical variables were considered. The Area under the curve (AUC) is the performance evaluation metric. Algorithms were validated against the TIMI and tested using the same validation data. SVM variable importance with backward elimination was used to select and rank important variables.ResultsDL algorithms outperform SVM and TIMI on the validation dataset; i) DL with SVM selected predictors (15 predictors, AUC = 0.97), ii) DL with expert-recommended predictors (16 predictors, AUC = 0.96), iii) SVM with selected predictors (15 predictors, AUC = 0.92), iv) SVM with expert-recommended predictors (AUC = 0.89) and TIMI (AUC = 0.82). Common predictors across SVM feature selection, expert-recommendation and TIMI are: age, heart rate, Killip class, fasting blood glucose, systolic blood pressure, comorbid diseases and ST-elevation. SVM feature selection also identified diuretics, PCI and pharmacotherapy drugs as predictors that improve mortality prediction in STEMI patients. Our findings suggest that the TIMI score underestimates patients risk of mortality. DL algorithm using selected predictors classified 35% of nonsurvival patients as high risk (risk probabilities >50%) compared to only 12.7% nonsurvival patients by TIMI (score >5) (Figure below).ConclusionsIn the ASIAN population, patients with STEMI can be better classified using the DL algorithm compared to the ML and TIMI score. Combining ML feature selection with DL allows the identification of distinct factors in a unique ASIAN population for better mortality prediction than relying solely on an expert recommendation as it is a very subjective approach. Continuous validation on population-specific algorithms using DL and ML is needed before implementing in a real clinical setting.Funding AcknowledgementType of funding sources: Public grant(s) – National budget only. Main funding source(s): Technology Development Fund 1TIMI performance on validation setDL performance on validation set

This paper presents transient stability assessment of a large actual power system using the probabilistic neural network (PNN) with enhanced feature selection and extraction method. The investigated large power system is divided into five smaller areas depending on the coherency of the areas when subjected to disturbances. This is to reduce the amount of data sets collected for the respective areas. Transient stability of the power system is first determined based on the generator relative rotor angles obtained from time domain simulations carried out by considering three phase faults at different loading conditions. The data collected from the time domain simulations are then used as inputs to the PNN. An enhanced feature selection and extraction methods are then incorporated to reduce the input features to the PNN which is used as a classifier to determine whether the power system is stable or unstable. It can be concluded that the PNN with enhanced feature selection and extraction methods reduces the time taken to train the PNN without affecting the accuracy of the classification results.

Microarray gene expression data are often classified by cell line or tumor type to assist in the diagnosis and prediction of human cancer. While microarray analysis demonstrates potential, choosing the most suitable machine learning approach is crucial for accurate cancer diagnosis and prediction. In this beginner’s guide, we outline how to select an optimal machine learning model for cancer phenotype prediction by comparing various existing methods. This study used three well-known machine learning methods: linear discriminant analysis, support vector machines, and random forest. To assess prediction performance, several performance metrics were considered, including model prediction accuracy (AC), the area under the curve (AUC), F-measure, the receiver operating characteristic (ROC) curve, and the precision-recall curve (PRC). Microarray gene expression data from two cancer types, leukemia and colon cancer were analysed. A cross-validation process with 100 resampling iterations was implemented to compute average performance measures (APM), ensuring the reliability of the results. Findings emphasize the significance of selecting the right machine learning model for accurate predictions of new samples. The methods employed provided satisfactory results, validated by various APM for both leukemia and colon cancer datasets. Notably, the random forest classifier exhibited the best performance in cancer prediction. International Journal of Statistical Sciences, Vol. 25(2), November, 2025, pp 47-57

Digital imaging

Gene expression microarray data is one of the most popular for dianosis of cancer. However, the microarray data have thousands of genes and very few samples, it is crucial to develop techniques to effectively gene selection for analysis. So, dimension reduction is an important issue for analysis, of which principle component analysis (PCA) is one of the frequently used methods, and in the previous works, the top several principle components are selected for modeling according to the descending order of eigenvalues. While in this paper, we argue that not all the first features are useful, but features should be selected form all the components by feature selection methods. We demonstrate a framework for selecting good feature subsets from all the principle components, leading to enhance classifier accuracy rates on the gene expression microarray data. As a case study, we have considered PCA for dimension reduction, decesion tree algorithms (DT) for feature selection, and then Multi Layer Perceptron network (MLP) for classification. Experimental results illustrate that our proposed framework is effective to enhance classification accuracy rates.

This study aims to develop and validate a machine learning-based mortality risk prediction model for V-A ECMO patients to improve the precision of clinical decision-making. This multicenter retrospective cohort study included 280 patients receiving V-A ECMO from the Second Affiliated Hospital of Guangxi Medical University, Yulin First People’s Hospital, and the MIMIC-IV database. The data from the Second Affiliated Hospital of Guangxi Medical University and the MIMIC-IV database were merged and randomly divided in a 7:3 ratio into a training set and an internal validation set, respectively. The dataset from Yulin First People’s Hospital was reserved as an external validation cohort. The primary study outcome was defined as in-hospital mortality.Feature selection was conducted using Lasso regression, followed by the development of six machine learning models: Logistic Regression, Random Forest (RF), Deep Neural Network (DNN), Support Vector Machine (SVM), LightGBM, and CatBoost. Model performance was assessed using the Area Under the Curve (AUC), accuracy, sensitivity, specificity, and F1 score. Model validation was performed through calibration and decision curve analysis. Feature importance was evaluated using SHAP, and subgroup analysis was conducted to assess the model’s applicability across different clinical scenarios. In internal validation, the Logistic Regression model performed the best, with an AUC of 0.86 (95% CI: 0.77–0.93), accuracy of 0.76, sensitivity of 0.73, specificity of 0.79, and an F1 score of 0.73. It outperformed other models (RF: AUC = 0.79, DNN: AUC = 0.78, SVM: AUC = 0.76, LightGBM: AUC = 0.71, CatBoost: AUC = 0.77). External validation yielded consistent results, with the Logistic Regression model’s AUC at 0.75 (95% CI: 0.56–0.92), accuracy of 0.69, sensitivity of 0.64, specificity of 0.73, and an F1 score of 0.66. Calibration curve analysis revealed that the Logistic Regression model had the lowest Brier score (0.1496), indicating the most reliable predicted probabilities. Decision curve analysis demonstrated that the model provided the highest net benefit across most decision thresholds. SHAP analysis identified lactate, age, and albumin as key predictors of mortality, with lactate and age positively correlated, and albumin negatively correlated. Subgroup analysis revealed better performance in the cardiac arrest group (AUC = 0.81), non-sepsis group (AUC = 0.75), and non-diabetes group (AUC = 0.78). The Logistic Regression-based mortality risk prediction model for V-A ECMO patients demonstrated comparable or even favorable performance to more complex machine learning models, with the advantage of higher interpretability.By explicitly incorporating lactate, age, and albumin as the principal predictors, this model facilitates precise risk stratification and provides practical support for clinical decision-making in ECMO management.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-25423-4.

Innovative strategies are necessary to enhance prostate cancer diagnosis whilst reducing unnecessary and invasive repeat biopsies. This study aimed to determine the significant parameters affecting repeat prostate biopsy outcomes and develop an optimal machine learning algorithm for predicting positive repeat prostate biopsy results. We analysed data from 174 men who underwent repeated prostate biopsies between January 2008 and December 2022. Systematic multiple-core, ultrasound-targeted prostate biopsies were performed, each two samples from prostatic transitional zone and peripheral zone were obtained bilaterally. Clinical characteristics were collected, including patients' age, initial prostate volume, prostate-specific antigen (PSA) level, free PSA (fPSA)/PSA ratio, biopsy core numbers, pathological result; The time interval between first and latest prostate biopsy; Latest PSA level, fPSA/PSA ratio, biopsy core numbers; And final pathological diagnosis. Six feature selection methods, namely, variable ranking, correlation matrix, random forest regression, recursive feature elimination, cross-validation and forward selection, were employed to identify key influencing factors for repeat biopsy outcomes. Subsequently, the performance of seven machine learning algorithms, namely, multivariable logistic regression (LR), K-nearest neighbour search (KNN), support vector classification (SVC), decision tree (DT), random forest classifier (RF), naïve Bayes classifier (NBC) and gradient booster tree (GB), was assessed based on accuracy, misclassification, recall, specificity, precision and receiver operating characteristic (ROC) area under the curve (AUC). About 70% of patients were used as the training dataset, meanwhile remaining 30% as validation dataset. 52 were ultimately diagnosed with prostate cancer following the final pathological examination. The remaining 122 patients were negative. Amongst six feature selection methods, the variable ranking emerged as the most effective method for identifying the essential factors influencing repeat biopsy results. Amongst the machine learning algorithms, SVC demonstrated superior accuracy (0.7365), low recall rate (0.2500) and low misclassification rate (0.2093) for both patients with cancer and healthy individuals. Meanwhile, the ROC curve of SVC showed a relatively high AUC (0.6871). We developed an SVC-based machine learning algorithm for predicting positive repeat prostate biopsy results. Our analysis revealed that initial and latest prostate volumes, initial and latest PSA levels, latest fPSA/PSA ratio and age are significant factors for this model.

DNA Microarrays enable the detection of genetic changes attributable to cancer by simultaneously analyzing the expression of thousands of genes. However, the identification of most relevant genes from thousands of gene expressions available in each biological sample, for cancer classification pose a great challenge. Although researchers have applied BPSO based wrapper approaches to get most relevant genes prior to cancer classification, these approaches didn’t achieve good classification accuracy due to the premature convergence caused by local stagnation problem. This paper proposes an improved Binary Particle Swarm Optimization (iBPSO) to tackle these issues. The proposed iBPSO based wrapper is examined using Naive-Bayes (NB), k-Nearest Neighbor (kNN), and Support Vector Machines (SVM) classifiers with stratified 5-fold cross-validation. The proposed iBPSO exhibited its efficacy in terms of classification accuracy and the number of selected genes in comparison to standard BPSO on six benchmark cancer microarray datasets. Our proposed iBPSO also effectively escapes from local minima stagnation.

Early detection of bladder cancer (BC) remains a major clinical challenge due to the limitations of current diagnostic methods, which are often invasive, expensive, or insufficiently sensitive, particularly for early-stage disease. Metabolomics approaches, when integrated with machine learning (ML) techniques, offer a powerful platform for identifying novel, non-invasive biomarkers. In this study, urinary volatile organic compounds (VOCs) were analysed from 87 BC patients and 90 age- and sex-matched cancer-free controls using headspace solid-phase microextraction coupled with gas chromatography-mass spectrometry (HS-SPME/GC-MS). Of the 90 VOCs identified, 27 were selected and used to train five ML algorithms-random forest (RF), support vector machine (SVM), partial least squares-discriminant analysis (PLS-DA), extreme gradient boosting (XGBoost), and k-nearest neighbors (k-NN). Model performance was evaluated using cross-validation and an independent validation set, with metrics including area under the curve (AUC), sensitivity, specificity, and accuracy. RF achieved the highest performance using all 27 features (AUC=0.913; sensitivity, specificity, and accuracy=85%). After feature selection, an eight-VOC panel improved performance on the validation set (AUC=0.872; sensitivity=89%, specificity=92%, accuracy=91%). The panel included ketones, aldehydes, a short fatty alcohol, and a phenol compound-seven elevated in BC, and one (acetone) decreased. This panel outperformed FDA-approved urinary assays and closely matched the specificity of urine cytology. These findings underscore the promise of VOC-based urinary biomarkers, in combination with ML, for the non-invasive detection of BC. Further large-scale validation studies are essential to confirm diagnostic utility and enable clinical translation.

Machine learning-based receiver operating characteristic (ROC) curves for crisp and fuzzy classification of DNA microarrays in cancer research

MR-guided laser interstitial thermal therapy (MRgLITT) is associated with lower seizure-free outcome but better safety profile compared to open surgery. However, the predictors of seizure freedom following MRgLITT remain uncertain. This study aimed to use machine learning to predict seizure-free outcome following MRgLITT and to identify important predictors of seizure freedom in children with drug-resistant epilepsy. This multicenter study included children treated with MRgLITT for drug-resistant epilepsy at 13 epilepsy centers. The authors used clinical data, diagnostic investigations, and ablation features to predict seizure-free outcome at 1 year post-MRgLITT. Patients from 12 centers formed the training cohort, and patients in the remaining center formed the testing cohort. Five machine learning algorithms were developed on the training data by using 10-fold cross-validation, and model performance was measured on the testing cohort. The models were developed and tested on the complete feature set. Subsequently, 3 feature selection methods were used to identify important predictors. The authors then assessed performance of the parsimonious models based on these important variables. This study included 268 patients who underwent MRgLITT, of whom 44.4% had achieved seizure freedom at 1 year post-MRgLITT. A gradient-boosting machine algorithm using the complete feature set yielded the highest area under the curve (AUC) on the testing set (AUC 0.67 [95% CI 0.50-0.82], sensitivity 0.71 [95% CI 0.47-0.88], and specificity 0.66 [95% CI 0.50-0.81]). Logistic regression, random forest, support vector machine, and neural network yielded lower AUCs (0.58-0.63) compared to the gradient-boosting machine but the findings were not statistically significant (all p > 0.05). The 3 feature selection methods identified video-EEG concordance, lesion size, preoperative seizure frequency, and number of antiseizure medications as good prognostic features for predicting seizure freedom. The parsimonious models based on important features identified by univariate feature selection slightly improved model performance compared to the complete feature set. Understanding the predictors of seizure freedom after MRgLITT will assist with prognostication.

This study applies machine learning (ML) techniques to classify fertile [for porphyry Cu and (or) Au systems] and barren adakites using geochemical data from New Brunswick, Canada. It emphasizes that not all intrusive units, including adakites, are inherently fertile and should not be directly used as the heat source evidence layer in mineral prospectivity mapping without prior analysis. Adakites play a crucial role in mineral exploration by helping distinguish between fertile and barren intrusive units, which significantly influence ore-forming processes. A dataset of 99 fertile and 66 barren adakites was analyzed using seven ML models: support vector machine (SVM), neural network, random forest (RF), decision tree, AdaBoost, gradient boosting, and logistic regression. These models were applied to classify 829 adakite samples from around the world into fertile and barren categories, with performance evaluated using area under the curve (AUC), classification accuracy, F1 score, precision, recall, and Matthews correlation coefficient (MCC). SVM achieved the highest performance (AUC = 0.91), followed by gradient boosting (0.90) and RF (0.89). For model validation, 160 globally recognized fertile adakites were selected from the dataset based on well-documented fertility characteristics. Among the tested models, SVM demonstrated the highest classification accuracy (93.75%), underscoring its effectiveness in distinguishing fertile from barren adakites for mineral prospectivity mapping. Statistical analysis and feature selection identified middle rare earth elements (REEs), including Gd and Dy, with Hf, as key indicators of fertility. A comprehensive analysis of 1596 scatter plots, generated from 57 geochemical variables, was conducted using linear discriminant analysis (LDA) to determine the most effective variable pairs for distinguishing fertile and barren adakites. The most informative scatter plots featured element vs. element combinations (e.g., Ga vs. Dy, Ga vs. Gd, and Pr vs. Gd), followed by element vs. major oxide (e.g., Fe2O3T vs. Gd and Al2O3 vs. Hf) and ratio vs. element (e.g., La/Sm vs. Gd, Rb/Sr vs. Hf) plots, whereas major oxide vs. major oxide, ratio vs. ratio, and major oxide vs. ratio plots had limited discriminatory power.

Predicting traditional Chinese medicine constitutions in adults aged ≥ 65 years: A machine learning approach.

As climate change intensifies, the frequency and severity of waterlogging are expected to increase, necessitating a deeper understanding of the cucumber response to this stress. In this study, three public RNA-seq datasets (PRJNA799460, PRJNA844418, and PRJNA678740) comprising 36 samples were analyzed. Various feature selection algorithms including Uncertainty, Relief, SVM (Support Vector Machine), Correlation, and logistic least absolute shrinkage, and selection operator (LASSO) were performed to identify the most significant genes related to the waterlogging stress response. These feature selection techniques, which have different characteristics, were used to reduce the complexity of the data and thereby identify the most significant genes related to the waterlogging stress response. Uncertainty, Relief, SVM, Correlation, and LASSO identified 4, 4, 10, 21, and 13 genes, respectively. Differential gene correlation analysis (DGCA) focusing on the 36 selected genes identified changes in correlation patterns between the selected genes under waterlogged versus control conditions, providing deeper insights into the regulatory networks and interactions among the selected genes. DGCA revealed significant changes in the correlation of 13 genes between control and waterlogging conditions. Finally, we validated 13 genes using the Random Forest (RF) classifier, which achieved 100% accuracy and a 1.0 Area Under the Curve (AUC) score. The SHapley Additive exPlanations (SHAP) values clearly showed the significant impact of LOC101209599, LOC101217277, and LOC101216320 on the model’s predictive power. In addition, we employed the Boruta as a wrapper feature selection method to further validate our gene selection strategy. Eight of the 13 genes were common across the four feature weighting algorithms, LASSO, DGCA, and Boruta, underscoring the robustness and reliability of our gene selection strategy. Notably, the genes LOC101209599, LOC101217277, and LOC101216320 were among genes identified by multiple feature selection methods from different categories (filtering, wrapper, and embedded). Pathways associated with these specific genes play a pivotal role in regulating stress tolerance, root development, nutrient absorption, sugar metabolism, gene expression, protein degradation, and calcium signaling. These intricate regulatory mechanisms are crucial for cucumbers to adapt effectively to waterlogging conditions. These findings provide valuable insights for uncovering targets in breeding new cucumber varieties with enhanced stress tolerance.