Performance Of Classifier Research Articles

Enhancement of Classifier Performance with Adam and RanAdam Hyper-Parameter Tuning for Lung Cancer Detection from Microarray Data-In Pursuit of Precision.

Microarray gene expression analysis is a powerful technique used in cancer classification and research to identify and understand gene expression patterns that can differentiate between different cancer types, subtypes, and stages. However, microarray databases are highly redundant, inherently nonlinear, and noisy. Therefore, extracting meaningful information from such a huge database is a challenging one. The paper adopts the Fast Fourier Transform (FFT) and Mixture Model (MM) for dimensionality reduction and utilises the Dragonfly optimisation algorithm as the feature selection technique. The classifiers employed in this research are Nonlinear Regression, Naïve Bayes, Decision Tree, Random Forest and SVM (RBF). The classifiers' performances are analysed with and without feature selection methods. Finally, Adaptive Moment Estimation (Adam) and Random Adaptive Moment Estimation (RanAdam) hyper-parameter tuning techniques are used as improvisation techniques for classifiers. The SVM (RBF) classifier with the Fast Fourier Transform Dimensionality Reduction method and Dragonfly feature selection achieved the highest accuracy of 98.343% with RanAdam hyper-parameter tuning compared to other classifiers.

Evaluation of the Relation between Ictal EEG Features and XAI Explanations.

Epilepsy is a neurological disease with one of the highest rates of incidence worldwide. Although EEG is a crucial tool for its diagnosis, the manual detection of epileptic seizures is time consuming. Automated methods are needed to streamline this process; although there are already several works that have achieved this, the process by which it is executed remains a black box that prevents understanding of the ways in which machine learning algorithms make their decisions. A state-of-the-art deep learning model for seizure detection and three EEG databases were chosen for this study. The developed models were trained and evaluated under different conditions (i.e., three distinct levels of overlap among the chosen EEG data windows). The classifiers with the best performance were selected, then Shapley Additive Explanations (SHAPs) and Local Interpretable Model-Agnostic Explanations (LIMEs) were employed to estimate the importance value of each EEG channel and the Spearman's rank correlation coefficient was computed between the EEG features of epileptic signals and the importance values. The results show that the database and training conditions may affect a classifier's performance. The most significant accuracy rates were 0.84, 0.73, and 0.64 for the CHB-MIT, Siena, and TUSZ EEG datasets, respectively. In addition, most EEG features displayed negligible or low correlation with the importance values. Finally, it was concluded that a correlation between the EEG features and the importance values (generated by SHAP and LIME) may have been absent even for the high-performance models.

On Unsupervised Domain Adaptation: Pseudo Label Guided Mixup for Adversarial Prompt Tuning

To date, a backbone of methods for unsupervised domain adaptation (UDA) involves learning label-discriminative features via a label classifier and domain-invariant features through a domain discriminator in an adversarial scheme. However, these methods lack explicit control for aligning the source data and target data within the same label class, degrading the classifier's performance in the target domain. In this paper, we propose PL-Mix, a pseudo label guided Mixup method based on adversarial prompt tuning. Specifically, our PL-Mix facilitates class-dependent alignment and can alleviate the impact of noisy pseudo-labels. We then theoretically justify that PL-Mix can improve the generalization for UDA. Extensive experiments of the comparison with existing models also demonstrate the effectiveness of PL-Mix.

Predictive Analysis for Supply Chain Management Using Extreme Gradient Boost Classifier

A novel approach to predictive analysis and demand forecasting within supply chain management, employing the Extreme Gradient Boosting (XGBoost) classifier. In response to the escalating complexity and volatility of supply chains, accurate demand forecasting is imperative for optimizing inventory, production scheduling, and overall operational efficiency. Traditional forecasting methods often struggle to capture the nonlinear relationships and intricate patterns inherent in supply chain data. Conversely, XGBoost offers a potent machine learning technique adept at handling nonlinear relationships and delivering robust predictions. The proposed framework involves data preprocessing, feature engineering, model training, and validation stages. Through a case study employing real-world supply chain data, we demonstrate the superior performance of the XGBoost classifier over traditional methods in terms of accuracy, robustness, and scalability. This study underscores XGBoost's potential as a valuable tool for demand forecasting in supply chain management, facilitating informed decision-making, optimized inventory management, cost reduction, and enhanced customer satisfaction. Furthermore, the framework's adaptability and extendibility make it applicable to diverse industries and domains, contributing to the advancement of supply chain management through the integration of machine learning techniques for more precise and efficient demand forecasting

Classification of Liver Disorders Diagnosis using Naïve Bayes Method

Liver diseases pose a significant health challenge, necessitating robust predictive tools for early diagnosis. This study aims to determine the predictive performance of Naive Bayes classifier, one of the data mining algorithms, in the classification of liver diseases. The study applied 5, 10 and 20-fold cross-validation method. Trying to determine the effect of the cross-validation (CV) method used on the classification performance, this study used the "BUPA" dataset in the UCI Machine Learning Repository database for this purpose. The dataset consists of 6 variables and 345 examples. Orange program was used for data analysis. The study showed that the accuracy of the Naive bayes method were 64.6%, 66.7% and 64.3%, respectively. Accordingly, it can be said that the 10-fold CV method performs better. Compared to similar studies, it can be claimed that the analysis results obtained with the Orange program are better.

Boosting Few-Shot Learning via Attentive Feature Regularization

Few-shot learning (FSL) based on manifold regularization aims to improve the recognition capacity of novel objects with limited training samples by mixing two samples from different categories with a blending factor. However, this mixing operation weakens the feature representation due to the linear interpolation and the overlooking of the importance of specific channels. To solve these issues, this paper proposes attentive feature regularization (AFR) which aims to improve the feature representativeness and discriminability. In our approach, we first calculate the relations between different categories of semantic labels to pick out the related features used for regularization. Then, we design two attention-based calculations at both the instance and channel levels. These calculations enable the regularization procedure to focus on two crucial aspects: the feature complementarity through adaptive interpolation in related categories and the emphasis on specific feature channels. Finally, we combine these regularization strategies to significantly improve the classifier performance. Empirical studies on several popular FSL benchmarks demonstrate the effectiveness of AFR, which improves the recognition accuracy of novel categories without the need to retrain any feature extractor, especially in the 1-shot setting. Furthermore, the proposed AFR can seamlessly integrate into other FSL methods to improve classification performance.

Decoding the writing styles of disciplines: A large-scale quantitative analysis

Disciplinary writing style stems from the practice of science, reflecting the scientific culture. This study aims to explore the differences and evolution of scientific writing styles from the perspective of disciplines. A large-scale quantitative analysis was conducted over 14 million abstracts from the Microsoft Academic Graph (MAG) database across eight soft and hard disciplines. Represented by a comprehensive set of 14 symbolic, lexical, syntactic, structural, and readability features, the evolution of disciplinary writing styles was analyzed over 30 years. Interpretable machine learning methods were performed to test the discernibility of writing styles across disciplines and disclose their linguistic differences. Our findings reveal the linguistic features of soft disciplines (Art, Philosophy, and Sociology) and Mathematics generally keep stabilized, and a general trend of increasing linguistic complexity was observed for Biology, Chemistry, Computer Science, and Psychology. The good performance of the pairwise writing style classifiers indicates a well discriminability of the writing styles between disciplines. A correlation between the performance of classifiers and the distance between disciplines was identified. The feature contribution analysis using SHapley Additive exPlanations (SHAP) and Kendall's Tau rank correlation revealed the detailed commonalities and disparities in disciplines’ linguistic features. This study provides profound insights into the understanding of scientific writing and norms, which further helps develop useful tools for academic text analysis, foster interdisciplinary communication, and assist educators to construct discipline-specific writing guidance.

Abstract 466: microRNA regulators of neuroendocrine differentiation in metastatic castration-resistant prostate cancer

Abstract Therapy-induced neuroendocrine prostate cancer (NEPC), an extremely aggressive variant of castration-resistant prostate cancer (CRPC), is increasing in incidence with the widespread use of highly potent androgen receptor (AR)-pathway inhibitors (APIs) via a reversible trans-differentiation process, referred to as neuroendocrine differentiation (NED). The molecular basis of NED is not completely understood leading to a lack of effective molecular markers for its diagnosis. We hypothesized that alterations in key miRNAs drive NED, leading to lineage switching and emergence of NEPC. The aims of our study are: (i) identify key miRNA alterations that occur as PCa cells undergo neuroendocrine differentiation; (ii) develop a miRNA-based classifier for predicting/diagnosing NEPC; (iii) understand the functional role of key miRNA alterations. To systematically investigate the key miRNAs alterations driving therapy induced NED, we performed small RNA-NGS in a retrospective cohort of human metastatic CRPC clinical samples + PDX models with adenocarcinoma features (CRPC-adeno) vs those with neuroendocrine features (CRPC-NE). Further, with the application of machine learning algorithms to sequencing data, we trained a ‘miRNA classifier’ that could robustly classify ‘CRPC-NE’ from ‘CRPC-Adeno’ cases. The performance of classifier was validated in additional cohorts of mCRPC patients acquired from independent sites by employing single model prediction algorithm. We further sought to understand the functional role of these miRNAs by overexpessing in PCa cell lines followed by functional assays. In vivo xenograft studies were performed using PC3 cells overexpressing miR-28/control miRNA. We demonstrate for the first time, that lineage switching to NE states is accompanied by key miRNA alterations. Our analysis, measured using receiver operating characteristic (ROC) curves, demonstrated an area under the curve (AUC) value of 0.8318, confirming the robustness of our classifier in distinguishing CRPC-NE from CRPC-Adeno. Furthermore, we identified a select group of 5 miRNAs that were pivotal in distinguishing between these two prostate cancer subtypes. We focused on miR-28-3p, little studied miRNA in prostate cancer. Our in vitro and in vivo data with prostate cancer cell lines and xenograft model suggests that miR-28-3p acts as a tumor suppressor miRNA in advanced stage prostate cancer. We further found mesenchymal gene Vimentin to be a direct miR-28-3p target. We propose that loss of miR-28-3p promotes epithelial-to-mesenchymal transition (EMT), thereby favoring NED. Our study has important clinical implications and transformative potential as our 'miRNA classifier' can be used as a molecular tool to stratify mCRPC patients into those with/without NED and guide treatment decisions. Further, we identify novel miRNA NED drivers that can be exploited for NEPC therapeutic targeting. Citation Format: Diana Asante, Sandip Nathani, Tae Jin Lee, Ashok Sharma, Sharanjot Saini. microRNA regulators of neuroendocrine differentiation in metastatic castration-resistant prostate cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 466.

Abstract 1056: Development of a B-cell epitope classifier for early detection of renal cell carcinoma

Abstract A blood biomarker for early detection of renal cell carcinoma (RCC) would help decrease the burden of 15,000 deaths annually in the U.S. (1). Antibodies against tumor (neo)antigens are thought to arise early in cancer development and might serve as a more sensitive biomarker for RCC than tumor-derived biomolecules. Serum Epitope Repertoire Analysis (SERA) of human serum/plasma uses a random bacterial peptide display library and next generation sequencing to allow for unbiased highly multiplexed profiling of B cell epitopes. In combination with bioinformatic methods and machine learning, a classifier was trained to predict the presence of RCC using a cohort of 163 treatment-naïve patients (median age 62, stage: I 35%, II 5%, III 20%, IV 40%) with clear cell RCC (ccRCC) and 438 age-matched healthy controls (median age 61). Classifier performance was evaluated by sensitivity, specificity, and receiver operator curve (ROC) area under the curve (AUC) analysis in an independent validation cohort of 43 treatment-naïve patients with ccRCC (age 22-86, median 63, stage: I 33%, II 2%, III 26%, IV 39%) and 1,759 healthy controls (age 18-89, median 40). A total of 1,465 individual B cell epitope features were discovered and enriched in ccRCC vs. controls and included in a random forest model. In the validation cohort, the ROC-AUC was 0.65, and sensitivities of 9%, 16%, 30%, and 36% were achieved at specificities of 99%, 98%, 95%, and 90%, respectively. Performance was better in both early stage ccRCC and low tumor grade (grade 1 or 2). For stage I, the ROC-AUC was 0.88, and sensitivities of 7%, 14%, 36%, and 57% were achieved at specificities of 99%, 98%, 95%, and 90%, respectively. For low tumor grade, the ROC-AUC was 0.86, and sensitivities of 10%, 20%, 50%, and 60% of were achieved at specificities of 99%, 98%, 95%, and 90%, respectively. The classifier did not miscategorize benign renal lesions as ccRCC (Mann-Whitney p-value 0.52, ROC-AUC 0.54) in a separate cohort of 23 patients with benign renal tumors vs. the 1,759 healthy controls. This preliminary performance for a B cell epitope-based classifier is promising: the better performance in early stage ccRCC suggests that B cell epitope biomarkers may complement the lower sensitivity of cell-free DNA for detecting early-stage cancer (2). Future work will further validate these results in larger multi-institutional cohorts to improve classification performance and explore if B cell epitope biomarkers can be combined with existing assays to develop an early detection composite biomarker for RCC and tumor grade. (1)Siegel RL, Miller KD, Wagle NS, Jemal A. Cancer statistics, 2023. CA Cancer J Clin. 2023 Jan;73(1):17-48.(2)Francini, E.; Fanelli, G.N.; Pederzoli, F.; Spisak, S.; Minonne, E.; Raffo, M.; Pakula, H.; Tisza, V.; Scatena, C.; Naccarato, A.G.; et al. Circulating Cell-Free DNA in Renal Cell Carcinoma: The New Era of Precision Medicine. Cancers 2022, 14, 4359. Citation Format: Thomas W. Campbell, Christian R. Hoerner, Elizabeth Alli, John T. Leppert, John C. Shon, Alice C. Fan. Development of a B-cell epitope classifier for early detection of renal cell carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1056.

Abstract 4414: Genome-wide DNA methylation classifier diagnoses pancreatic ductal adenocarcinoma in CUP setting

Abstract Introduction: The metastasis of a pancreatic ductal adenocarcinoma (PAAD) is a diagnosis of exclusion and one of the most common causes of cancer of unknown primary (CUP). We have recently developed a genome-wide DNA methylation-based neural network classifier that can accurately differentiate between liver metastasis of a PAAD and intrahepatic cholangiocarcinoma (iCCA) (PAAD-iCCA-Classifier). Therefore, the aim of our study was to test whether our PAAD-iCCA-Classifier can be extended to be used to correctly diagnose PAAD metastases from other sites in CUP setting. Methods: For this purpose, we enhanced the anomaly detection layer of the classifier by incorporating ten mimicker carcinomas to be excluded by this layer. We used a validation set 1 (n=3786) including primary PAAD (n=242), PAAD liver metastases (n=20), iCCA (n=151) and 10 other mimicker carcinomas (n=3373) and a validation set 2 (n=26) including primary PAAD from a real-life clinical cohort from an independent institution to validate the classifier. Next, we tested the classifier on 16 PAAD initially considered CUP samples (test set) from different sites: peritoneum, lung, liver, and lymph node. The clinical history and diagnostic imaging of these samples were used to confirm PAAD as the most probable origin. We further performed differentially methylated probes (DMP) and copy number alterations (CNA) analysis of primary PAAD and metastatic PAAD from different locations. Results: The improved version of the PAAD-iCCA-Classifier achieved an accuracy of 98.43% on the validation set 1 and was able to exclude most of the mimicker carcinomas. On validation set 2, the classifier achieved an accuracy of 88.46%. Medical history, imaging and immunohistochemical analysis of the test set samples confirmed the diagnosis of PAAD. The DNA methylation classifier correctly labeled 15/16 PAAD metastatic samples as PAAD (93.75% accuracy). We observed that the classifier performance was negatively affected by a high CD3+ immune infiltrate and positively affected by high tumor purity and high proliferation rate. CNA revealed that PAAD liver metastases have a distinct CNA profile characterized by chromosome 6, 9 (CDKN2A/B) and 18q (SMAD4) deletions. DMP analysis showed that PAAD liver metastases have global hypomethylation of both promoter- and enhancer-associated CpGs compared to primary PAAD and PAAD peritoneal carcinomatosis. Finally, gene ontology analysis revealed that different epithelial-mesenchymal transition pathways are activated in PAAD liver metastases compared to PAAD peritoneal carcinomatosis. Conclusion: Our tool performs well in classifying metastatic PAAD samples and could be of great clinical use when a PAAD origin is suspected in the case of a CUP. DMP and CNA profiles show that PAAD liver metastases may have a distinct DNA methylation and copy number profile compared to primary and peritoneal carcinomatosis PAAD. Citation Format: Teodor G. Calina, Eilís Perez, Simon Schallenberg, Horst David, Erik Knutsen, David Capper, Mihnea P. Dragomir. Genome-wide DNA methylation classifier diagnoses pancreatic ductal adenocarcinoma in CUP setting [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4414.

Abstract 1040: Development of a novel RNA-based microsatellite stable predictive response signature (MSS-PRS) to identify MSS colorectal cancer (CRC) patients with a microsatellite instability-high (MSI-H) molecular phenotype

Abstract Introduction: MSI-H and mismatch repair deficient (dMMR) tumors are associated with favorable immune checkpoint inhibitor (ICI) responses (PMID: 3326454). However, up to 95% of CRC tumors are MSS/proficient MMR (pMMR) leading to poorer prognosis and treatment outcomes than MSI-H/dMMR patients (PMID: 12867608, 26028255). In MSS/pMMR metastatic CRC, multiple combination therapies are being investigated despite a lack of biomarkers to guide therapy. Therefore, we developed an RNA-based MSS-PRS with a primary objective to identify MSS tumors that have MSI-H/dMMR molecular features indicative of ICI response. Experimental Procedures: The MSS-PRS was developed using the TCGA CRC cohort (COAD; n=268). Training labels were assigned using the MSI Mantis score and mutation status from 8 genes with reported MSI association. Samples called “activated” had both a gene mutation and an MSI Mantis score > 0.4 (i.e., MSI-H), and samples called “non-activated” were wild type for all genes and had MSI Mantis score < 0.4. All other samples were designated “ambiguous” and excluded from training. Two thirds of the non-ambiguous samples were assigned to the classifier training set and all remaining samples were assigned to the test set. Using ClaNC software (PMID: 16269418) and cross-validation in the training set, a nearest centroid classifier was developed from a set of high mean, high variance candidate genes to select an optimal gene set to separate activated and non-activated groups. The classifier performance was evaluated in the test set. Results: The MSS-PRS contained 112-genes enriched for mismatched DNA binding, DNA damage repair, PD-L1, innate and adaptive immune response, cellular immunity, and cytokines. In the test set, the classifier called 44 of 46 MSI-H samples activated; of the remaining 77 samples that were MSS tumors, 38 were called activated and 39 were called not activated. Visual inspection of heat maps of tumor by MSI/MSS-PRS status and their association with TMB and CRC-related mutations suggest greater differences between MSS tumors called activated and MSI-H tumors than between MSS tumors called activated and MSS tumors called not activated. In contrast, immune marker expression profiles in MSS tumors called activated were markedly more similar to MSI-H tumors compared to MSS tumors called non-activated. Summary and Conclusions: Herein we described the development of a novel MSS-PRS that captures an MSI-H/dMMR-like molecular phenotype in a subset of patients with MSS tumors despite their lack of high TMB or overt MSI defects. Based upon these initial findings, further development of the MSS-PRS and its clinical validation as a tool to select patients with MSS tumors who may benefit from ICI-containing treatment regimens is warranted. Citation Format: Kirk Pappan, Gregory Mayhew, Jonathan Shepherd, Yuelong Guo, Kirk Beebe, Joel Eisner, Michael Milburn. Development of a novel RNA-based microsatellite stable predictive response signature (MSS-PRS) to identify MSS colorectal cancer (CRC) patients with a microsatellite instability-high (MSI-H) molecular phenotype [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1040.

A comparative analysis of machine learning classifiers in the classification of resonant asteroids

This study explores how well various machine learning classifiers can identify mean-motion resonances in the main belt using supervised learning. The most popular classifiers are assessed: k-Nearest Neighbours, Decision Tree, Gradient Boosting, AdaBoost, Random Forest, and Naïve Bayes. In contrast to previous studies that often relied on default ML configurations, this research conducts a detailed investigation, fine-tuning, and testing of each classifier across various parameters. The results show that simpler models, especially k-Nearest Neighbours and Decision Tree, perform better than more complex ones, particularly in terms of F1 scores. The paper provides guides on selecting features, parameters, and training set sizes for optimal classifier performance and outlines a method for developing effective machine-learning models for asteroid classification.

Long-term prediction by DNA methylation of high-grade cervical intraepithelial neoplasia: Results of the ARTISTIC cohort.

Methylation markers have shown potential for triaging high-risk HPV-positive (hrHPV+) women to identify those at increased risk of invasive cervical cancer (ICC). Our aim was to assess the performance of the S5 DNA methylation classifier for predicting incident high-grade cervical intraepithelial neoplasia (CIN) and ICC among hrHPV+ women in the ARTISTIC screening trial cohort. The S5 classifier, comprising target regions of tumour suppressor gene EPB41L3 and L1 and L2 regions of HPV16, HPV18, HPV31, and HPV33, was assayed by pyrosequencing in archived hrHPV+ liquid-based samples from 343 women with high-grade disease (139 CIN2, 186 CIN3, and 18 ICC) compared to 800 hrHPV+ controls. S5 DNA methylation correlated directly with increasing severity of disease and inversely with lead time to diagnosis. S5 could discriminate between hrHPV+ women who developed CIN3 or ICC and hrHPV+ controls (p <.0001) using samples taken on average 5 years before diagnosis. This relationship was independent of cytology at baseline. The S5 test showed much higher sensitivity than HPV16/18 genotyping for identifying prevalent CIN3 (93% vs. 61%, p = .01) but lower specificity (50% vs. 66%, p <.0001). The S5 classifier identified most women at high risk of developing precancer and missed very few prevalent advanced lesions thus appearing to be an objective test for triage of hrHPV+ women. The combination of methylation of host and HPV genes enables S5 to combine the predictive power of methylation with HPV genotyping to identify hrHPV-positive women who are at highest risk of developing CIN3 and ICC in the future.

Integrated Approach for ECG Signal Classification Using Fused Slantlet Transform and Autoregressive Features

Sudden cardiac death remains a significant concern, emphasizing the critical need for intelligent diagnostic techniques in cardiac disease detection. Cardiologic Vascular Diseases (CVD) pose a significant global health challenge, often associated with unhealthy lifestyle habits. Electrocardiogram (ECG) monitoring plays a crucial role in assessing heart function, but its interpretation can be challenging due to signal complexity. This study proposes a novel approach utilizing fused Slantlet-Spatial features alongside Auto Regression (AR) coefficients to classify electrocardiogram (ECG) signals. Leveraging the Slantlet Transform (SLT) for its superior time localization capabilities and the simplicity of AR coefficients, this methodology aims to enhance the accuracy of cardiac illness classification. Nineteen distinct features were extracted from 300 ECG signals, each comprising 512 samples, enriching the classification scheme's efficacy. The fused feature extraction technique demonstrated success in improving overall accuracy. Employing Binary Particle Swarm Optimization (BPSO), irrelevant features were pruned, resulting in a refined feature vector of only 14 relevant features. This reduction in feature dimensionality significantly influenced the classification accuracy. Subsequently, the refined feature vector was introduced to the classification stage, employing four distinct classifiers: Support Vector Machine (SVM) with various kernel functions, K-Nearest Neighbors (K-NN), Naïve Bayes (NB), and Decision Tree (DT). Simulation results revealed the feasibility of achieving an average classification accuracy of 99.67% across ECG signal categories. Notably, the polynomial kernel function-SVM classifier, with only 2-fold cross-validation, yielded optimal performance. This study emphasizes the potential of fused Slantlet-Spatial and AR features in enhancing the accuracy of cardiac disease classification. Moreover, the efficacy of BPSO in feature selection and the superior performance of the SVM classifier highlight promising avenues for further research in intelligent cardiac diagnosis systems. The proposed approach outperforms existing methods and contributes to the early detection of cardiovascular abnormalities, potentially improving patient outcomes.

The Sound of Surveillance: Enhancing Machine Learning-Driven Drone Detection with Advanced Acoustic Augmentation

In response to the growing challenges in drone security and airspace management, this study introduces an advanced drone classifier, capable of detecting and categorizing Unmanned Aerial Vehicles (UAVs) based on acoustic signatures. Utilizing a comprehensive database of drone sounds across EU-defined classes (C0 to C3), this research leverages machine learning (ML) techniques for effective UAV identification. The study primarily focuses on the impact of data augmentation methods—pitch shifting, time delays, harmonic distortion, and ambient noise integration—on classifier performance. These techniques aim to mimic real-world acoustic variations, thus enhancing the classifier’s robustness and practical applicability. Results indicate that moderate levels of augmentation significantly improve classification accuracy. However, excessive application of these methods can negatively affect performance. The study concludes that sophisticated acoustic data augmentation can substantially enhance ML-driven drone detection, providing a versatile and efficient tool for managing drone-related security risks. This research contributes to UAV detection technology, presenting a model that not only identifies but also categorizes drones, underscoring its potential for diverse operational environments.

An enhanced mechanism for detection of spam emails by deep learning technique with bio-inspired algorithm

This study aims to create a highly effective system for classifying email spam, with the key objective of improving performance and accuracy in classification. Rigorous pre-processing techniques, including lemmatization and tokenization, are applied to refine the dataset. The impact of optimal feature selection on deep learning algorithm stability with varying training datasets is investigated and to improve classifier performance, pre-stages such as feature extraction and selection using bag of words are utilized. The hybrid model is further enhanced by combining feature extraction and classification algorithms with optimization. The inclusion of multiple features, such as n-gram features, addresses the 'Curse of Dimensionality,' and optimal feature selection is employed to improve the spam detection process. The proposed system undergoes three main modifications: multiple feature extraction, feature selection, and an enhanced hybrid model, all aimed at improving spam detection accuracy. The e-mail classification phase involves mapping between training and testing sets, with deep learning algorithms utilized for feature extraction and classification.The recently introduced Sand Cat Swarm Optimization algorithm is employed to refine weights during each epoch, enhancing accuracy and minimizing loss. The proposed system extends its capabilities to identify phishing attacks within the network, offering a comprehensive approach to email security.

Highly Imbalanced Classification of Gout Using Data Resampling and Ensemble Method

Gout is one of the most painful diseases in the world. Accurate classification of gout is crucial for diagnosis and treatment which can potentially save lives. However, the current methods for classifying gout periods have demonstrated poor performance and have received little attention. This is due to a significant data imbalance problem that affects the learning attention for the majority and minority classes. To overcome this problem, a resampling method called ENaNSMOTE-Tomek link is proposed. It uses extended natural neighbors to generate samples that fall within the minority class and then applies the Tomek link technique to eliminate instances that contribute to noise. The model combines the ensemble ’bagging’ technique with the proposed resampling technique to improve the quality of generated samples. The performance of individual classifiers and hybrid models on an imbalanced gout dataset taken from the electronic medical records of a hospital is evaluated. The results of the classification demonstrate that the proposed strategy is more accurate than some imbalanced gout diagnosis techniques, with an accuracy of 80.87% and an AUC of 87.10%. This indicates that the proposed algorithm can alleviate the problems caused by imbalanced gout data and help experts better diagnose their patients.

Performance assessment of machine learning algorithms for mapping of land use/land cover using remote sensing data

The rapid increase in population accelerates the rate of change of Land use/Land cover (LULC) in various parts of the world. This phenomenon caused a huge strain for natural resources. Hence, continues monitoring of LULC changes gained a significant importance for management of natural resources and assessing the climate change impacts. Recently, application of machine learning algorithms on RS (remote sensing) data for rapid and accurate mapping of LULC gained significant importance due to growing need of LULC estimation for ecosystem services, natural resource management and environmental management. Hence, it is crucial to access and compare the performance of different machine learning classifiers for accurate mapping of LULC. The primary objective of this study was to compare the performance of CART (Classification and Regression Tree), RF (Random Forest) and SVM (Support Vector Machine) for LULC estimation by processing RS data on Google Earth Engine (GEE). In total four classes of LULC (Water Bodies, Vegetation Cover, Urban Land and Barren Land) for city of Lahore were extracted using satellite images from Landsat-7, Landsat-8 and Landsat-9 for years 2008, 2015 and 2022, respectively. According to results, RF is the best performing classifier with maximum overall accuracy of 95.2% and highest Kappa coefficient value of 0.87, SVM achieved maximum accuracy of 89.8% with highest Kappa of 0.84 and CART showed maximum overall accuracy of 89.7% with Kappa value of 0.79. Results from this study can give assistance for decision makers, planners and RS experts to choose a suitable machine learning algorithm for LULC classification in an unplanned urbanized city like Lahore.

Deep convolutional neural networks with genetic algorithm-based synthetic minority over-sampling technique for improved imbalanced data classification

Imbalanced data classification presents a challenge in machine learning, inducing biased model learning. Moreover, data dimensionality poses another challenge as it highly impacts classifier performance. This paper proposes a new deep-learning method that combines feature selection with oversampling to address these challenges. The proposed approach, GA-SMOTE-DCNN, integrates a genetic algorithm (GA) for feature selection, SMOTE for oversampling, and a deep 1D-convolutional neural network (DCNN) for classification. This study reveals that pre-splitting the data into training and testing sets before applying SMOTE results in higher accuracy, showing an improvement in accuracy ranging between 1.94% and 3.98% compared to post-SMOTE splitting for each dataset. This method achieved accuracy rates of 86.81% for the Balance Scale dataset, 86.15% for the Oil Spill dataset, 89.21% for the Yeast dataset, 91.32% for the Mammography dataset, 88.23% for the Australian credit dataset, and 89.53% for the German Credit dataset when compared with benchmark methods, underscoring its significance in tackling high-dimensional and imbalanced data classification problems. This method demonstrates scalability in effectively addressing challenges associated with high-dimensional and imbalanced data classification across various domains.

Synthesizing class labels for highly imbalanced credit card fraud detection data

Acquiring labeled datasets often incurs substantial costs primarily due to the requirement of expert human intervention to produce accurate and reliable class labels. In the modern data landscape, an overwhelming proportion of newly generated data is unlabeled. This paradigm is especially evident in domains such as fraud detection and datasets for credit card fraud detection. These types of data have their own difficulties associated with being highly class imbalanced, which poses its own challenges to machine learning and classification. Our research addresses these challenges by extensively evaluating a novel methodology for synthesizing class labels for highly imbalanced credit card fraud data. The methodology uses an autoencoder as its underlying learner to effectively learn from dataset features to produce an error metric for use in creating new binary class labels. The methodology aims to automatically produce new labels with minimal expert input. These class labels are then used to train supervised classifiers for fraud detection. Our empirical results show that the synthesized labels are of high enough quality to produce classifiers that significantly outperform a baseline learner comparison when using area under the precision-recall curve (AUPRC). We also present results of varying levels of positive-labeled instances and their effect on classifier performance. Results show that AUPRC performance improves as more instances are labeled positive and belong to the minority class. Our methodology thereby effectively addresses the concerns of high class imbalance in machine learning by creating new and effective class labels.