Binary Classification Model Research Articles

Renal Pathological Image Classification Based on Contrastive and Transfer Learning

Following recent advancements in medical laboratory technology, the analysis of high-resolution renal pathological images has become increasingly important in the diagnosis and prognosis prediction of chronic nephritis. In particular, deep learning has been widely applied to computer-aided diagnosis, with an increasing number of models being used for the analysis of renal pathological images. The diversity of renal pathological images and the imbalance between data acquisition and annotation have placed a significant burden on pathologists trying to perform reliable and timely analysis. Transfer learning based on contrastive pretraining is emerging as a viable solution to this dilemma. By incorporating unlabeled positive pretraining images and a small number of labeled target images, a transfer learning model is proposed for high-accuracy renal pathological image classification tasks. The pretraining dataset used in this study includes 5000 mouse kidney pathological images from the Open TG-GATEs pathological image dataset (produced by the Toxicogenomics Informatics Project of the National Institutes of Biomedical Innovation, Health, and Nutrition in Japan). The transfer training dataset comprises 313 human immunoglobulin A (IgA) chronic nephritis images collected at Fukushima Medical University Hospital. The self-supervised contrastive learning algorithm “Bootstrap Your Own Latent” was adopted for pretraining a residual-network (ResNet)-50 backbone network to extract glomerulus feature expressions from the mouse kidney pathological images. The self-supervised pretrained weights were then used for transfer training on the labeled images of human IgA chronic nephritis pathology, culminating in a binary classification model for supervised learning. In four cross-validation experiments, the proposed model achieved an average classification accuracy of 92.2%, surpassing the 86.8% accuracy of the original RenNet-50 model. In conclusion, this approach successfully applied transfer learning through mouse renal pathological images to achieve high classification performance with human IgA renal pathological images.

Abstract LB185: High resolution telomere measurements in human cancer and aging using long-read nanopore sequencing

Abstract Telomere length is an important biomarker of cellular replicative potential and aging, but existing measurement methods are limited in resolution and accuracy. During human carcinogenesis, telomeres shorten profoundly in the earliest stages that precede tumor invasion. Telomere shortening can either halt cancer progression by inducing senescence and cell death or can promote cancer development by destabilizing chromosomes and generating chromosomal rearrangements. Stabilization of telomeres, typically by the upregulation of telomerase, is required for progression of nearly all human cancers. Here, we deploy digital telomere measurement by nanopore sequencing to understand how distributions of human telomere length change during aging and following carcinogenesis. We measure telomere attrition and de novo elongation with unprecedented resolution in genetically defined populations of human stem cells, in cells from the peripheral blood of healthy individuals and patients with genetic defects in telomere maintenance, cancer patients, and cancer cell lines. We find that human aging is accompanied by a progressive loss of long telomeres and an accumulation of shorter telomeres in white blood cells. In patients with inherited defects in telomere maintenance, the accumulation of short telomeres is more pronounced and correlates with phenotypic severity. We apply machine learning to train a binary classification model that distinguishes healthy individuals from those with telomere biology disorders. Next, we investigated the telomere lengths of a cohort of twenty colorectal tumors with patient-matched healthy colonic epithelia and found that two-thirds of colorectal tumors have telomere length distributions significantly shorter than the surrounding normal tissue. Finally, we explore the utility of telomere length distributions in cancer cell lines as clinically actionable disease biomarkers. This sequencing and bioinformatic pipeline will advance our understanding of telomere maintenance mechanisms and the use of telomere length as a clinical biomarker of aging and disease. Citation Format: Santiago Sanchez, Jessica Gu, Anudeep Golla, William Shomali, Annika Martin, Dirk Hockemeyer, Sharon Savage, Steven Artandi. High resolution telomere measurements in human cancer and aging using long-read nanopore sequencing [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB185.

Parental status and markers of brain and cellular age: A 3D convolutional network and classification study

Recent research shows prominent effects of pregnancy and the parenthood transition on structural brain characteristics in humans. Here, we present a comprehensive study of how parental status and number of children born/fathered links to markers of brain and cellular ageing in 36,323 UK Biobank participants (age range 44.57–82.06 years; 52% female). To assess global effects of parenting on the brain, we trained a 3D convolutional neural network on T1-weighted magnetic resonance images, and estimated brain age in a held-out test set. To investigate regional specificity, we extracted cortical and subcortical volumes using FreeSurfer, and ran hierarchical clustering to group regional volumes based on covariance. Leukocyte telomere length (LTL) derived from DNA was used as a marker of cellular ageing. We employed linear regression models to assess relationships between number of children, brain age, regional brain volumes, and LTL, and included interaction terms to probe sex differences in associations. Lastly, we used the brain measures and LTL as features in binary classification models, to determine if markers of brain and cellular ageing could predict parental status. The results showed associations between a greater number of children born/fathered and younger brain age in both females and males, with stronger effects observed in females. Volume-based analyses showed maternal effects in striatal and limbic regions, which were not evident in fathers. We found no evidence for associations between number of children and LTL. Classification of parental status showed an Area under the ROC Curve (AUC) of 0.57 for the brain age model, while the models using regional brain volumes and LTL as predictors showed AUCs of 0.52. Our findings align with previous population-based studies of middle- and older-aged parents, revealing subtle but significant associations between parental experience and neuroimaging-based surrogate markers of brain health. The findings further corroborate results from longitudinal cohort studies following parents across pregnancy and postpartum, potentially indicating that the parenthood transition is associated with long-term influences on brain health.

An Approach for Detecting Security Attacks using Machine Learning in IoT Environment

The strong security measures are becoming even more necessary to protect these networked systems as a result of the proliferation of Internet of Things (IoT) devices. The dynamic and varied nature of IoT networks frequently makes it impossible for traditional security solutions to be effective. The method suggested in this research uses machine learning to identify security attacks in IoT contexts. The suggested method makes use of the capabilities of machine learning algorithms to examine the enormous amounts of data produced by IoT devices. The system can develop the ability to recognise possible security threats and take immediate action by training models on labelled datasets that include both normal and attack patterns. In this paper multiple ML models for security threat detection in IoT environments in this study. Our evaluation uses both binary and multiclass classification models in an effort to accurately reflect the variety of assaults that can be found in IoT environments. The proposed method provide a new specialised IoT dataset that was created especially to reflect the characteristics of actual IoT environments in order to assure the validity of our assessment. By completing this extensive analysis, we hope to shed light on how well AI-based methods for security attack detection in IoT contexts work. The results of this study can help researchers and professionals decide which ML models and feature engineering techniques are best for IoT security. At the end of the day, we want to help with the creation of reliable security systems that safeguard IoT devices and shield user data from harmful attacks.

A Hybrid Perspective on Threat Analysis and Activity-Based Attack Modeling for Strengthening Access Control in IoT

The rapid expansion of Internet of Things (IoT) devices has resulted in an unparalleled surge in the production of data and interconnectivity. Nevertheless, as IoT ecosystems become increasingly intricate, security concerns become of utmost importance, particularly in access control systems. The objective of this research is to improve the security of IoT access control by utilizing a hybrid model for analyzing threats and modeling attacks based on activities. This study has two primary objectives: a) A hybrid classification model is used to predict labels (attack or not) in binary classification with an impressive accuracy of 98.18%. b) Another hybrid classification model is employed to predict types of attacks in M2M communication, achieving a commendable accuracy of 90%. The primary goal is to create and assess a hybrid classification model for binary classification. This model will differentiate between regular system behavior and malicious attacks on access control schemes in the Internet of Things (IoT). The hybrid model, which combines the strengths of Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM) networks, achieves an exceptional accuracy rate of 98.18%. The model's high accuracy demonstrates its effectiveness in precisely detecting potential threats and minimizing false positives, thereby establishing a strong basis for improving access control security. The second objective focuses on the complex area of security, with the goal of categorizing distinct forms of attacks in Machine-to-Machine (M2M) communication within the Internet of Things (IoT) framework. The hybrid classification model, employing both GRU and LSTM networks, achieves a remarkable accuracy of 90%. This accomplishment showcases the model's aptitude in detecting and distinguishing different types of attacks, including Distributed Denial of Service (DDoS) and Man-in-the-Middle attacks. The hybrid model provides security professionals with valuable insights to proactively respond to diverse threats in M2M communication by accurately classifying attack types. This strengthens the overall security posture of IoT access control systems. Overall, this study offers a thorough and efficient combination of threat analysis and activity-based attack modeling to enhance access control in IoT. The obtained accuracies in binary classification and prediction of attack types highlight the practical usability of the suggested hybrid model, establishing a strong basis for improving the security of IoT access control systems against evolving cyber threats

Detecting and quantifying zero tillage technology adoption in Indian smallholder systems using Sentinel-2 multi-spectral imagery

Zero tillage (ZT), an important component of Conservation Agriculture, has enormous potential to curb emissions from residue burning, increase soil organic carbon and water retention, reduce land preparation costs and increase the long-term productivity and profitability of the farming system. Despite the promise of ZT, little is known about how widely it has been adopted at regional scales in smallholder systems, where management is heterogeneous. Identifying ZT diffusion patterns across space and time along with other popular tillage technologies, such as conventional tillage (CT) and shallow tillage (ST), helps to target and disseminate the most effective technologies and estimate their climate change mitigation potential. Acknowledging the complexities involved in distinguishing ZT from CT and ST in smallholder fields, this study utilized an innovative two-step change detection method leveraging early-season Sentinel-2 multi-spectral imagery. We developed and applied our model in the Indian state of Punjab over three years (2020–2022). Our method outperformed traditional binary classification models, achieving 81 % accuracy. The analysis indicated that areas under different tillage types changed over time across Punjab. Specifically, from 2020 to 2021, we found a 33 % and 4 % decrease in ZT and CT, respectively. However, a 29 % increase is observed in CT adoption. On the other hand, from 2021 to 2022, the adoption rates for ZT and CT increased by 18 % and 2 %, respectively, while ST adoption decreased by 12 %. Overall, this study demonstrates the potential use of early season Sentinel-2 imagery to map the adoption of tillage practices in smallholder systems. Our approach can provide large-scale information on technology uptake, aiding policies to implement carbon markets and the scaling up of sustainable agricultural practices in India.

Analysis Of One-Class and Binary Classification For COVID-19 Image Data

With the recent Coronavirus Disease 2019 (COVID-19) outbreak, there has been a growing interest in the utilization of machine learning models to aid in medical diagnostics. However, during the initial phases of the pandemic, the available datasets were severely imbalanced due to a scarcity of positive cases. These datasets predominantly comprised normal cases, with only a limited number of disease instances. This inherent data imbalance restricted the effectiveness of binary classification, prompting an exploration of one-class classification. This study aims to compare the performance of one-class and binary classification models during the early stages of COVID-19 development. The objective is to provide healthcare professionals with the most effective machine learning-assisted solutions for handling future outbreaks of a similar nature. The experiment incorporates three distinct models: One-Class Support Vector Machine (OCSVM) for one-class classification, alongside Categorical Boosting (CatBoost) and Extreme Gradient Boosting (XGBoost) for binary classification. These models undergo a rigorous tuning process employing grid search, using training data composed of features extracted from compressed chest X-ray images. In summary, the research findings unequivocally demonstrate the superior performance of one-class classification over binary classification in all early stages, achieving an impressive F1 score of 0.46, compared to binary classification's modest score of 0.063. Consequently, it is strongly recommended to employ a one-class classification model to enhance real-life diagnostic processes. However, it is anticipated that as more positive cases are incorporated into the dataset, achieving a more balanced distribution, binary classification may exhibit greater potential. Thus, medical practitioners should reconsider the applicability of binary classification in later stages when the dataset attains improved balance.

Survival strategies for family-run homestays: analyzing user reviews through text mining

Online booking of homestays through e-travel portals is based on the virtual brand and perception, which are largely affected by user-generated electronic word-of-mouth (eWOM). With the objective of mining actionable insights from eWOM, this study conducted opinion mining for homestays located in four thematic areas of Kerala. Accordingly, various techniques have been deployed, such as sentiment and emotional analyses, topic modeling, and clustering methods. Key themes revealed from topic modeling were breakfast, facilities provided, ambience, bathroom, cleanliness, hospitality exhibited, and satisfaction with the host. A lasso logistic regression-based predictive binary text classification model (with 97.6% accuracy) for homestay recommendations was developed. Our findings and predictive model have implications for managers and homestay owners in devising appropriate marketing strategies and improving their overall guest experience.

Predicting resprouting of Platanus × hispanica following branch pruning by means of machine learning.

Resprouting is a crucial survival strategy following the loss of branches, being it by natural events or artificially by pruning. The resprouting prediction on a physiological basis is a highly complex approach. However, trained gardeners try to predict a tree's resprouting after pruning purely based on their empirical knowledge. In this study, we explore how far such predictions can also be made by machine learning. Table-topped annually pruned Platanus × hispanica trees at a nursery were LiDAR-scanned for two consecutive years. Topological structures for these trees were abstracted by cylinder fitting. Then, new shoots and trimmed branches were labelled on corresponding cylinders. Binary and multiclass classification models were tested for predicting the location and number of new sprouts. The accuracy for predicting whether having or not new shoots on each cylinder reaches 90.8% with the LGBMClassifier, the balanced accuracy is 80.3%. The accuracy for predicting the exact numbers of new shoots with the GaussianNB model is 82.1%, but its balanced accuracy is reduced to 42.9%. The results were validated with a separate dataset, proving the feasibility of resprouting prediction after pruning using this approach. Different tree species, tree forms, and other variables should be addressed in further research.

An Empirical Evaluation of a Novel Ensemble Deep Neural Network Model and Explainable AI for Accurate Segmentation and Classification of Ovarian Tumors Using CT Images.

Ovarian cancer is one of the leading causes of death worldwide among the female population. Early diagnosis is crucial for patient treatment. In this work, our main objective is to accurately detect and classify ovarian cancer. To achieve this, two datasets are considered: CT scan images of patients with cancer and those without, and biomarker (clinical parameters) data from all patients. We propose an ensemble deep neural network model and an ensemble machine learning model for the automatic binary classification of ovarian CT scan images and biomarker data. The proposed model incorporates four convolutional neural network models: VGG16, ResNet 152, Inception V3, and DenseNet 101, with transformers applied for feature extraction. These extracted features are fed into our proposed ensemble multi-layer perceptron model for classification. Preprocessing and CNN tuning techniques such as hyperparameter optimization, data augmentation, and fine-tuning are utilized during model training. Our ensemble model outperforms single classifiers and machine learning algorithms, achieving a mean accuracy of 98.96%, a precision of 97.44%, and an F1-score of 98.7%. We compared these results with those obtained using features extracted by the UNet model, followed by classification with our ensemble model. The transformer demonstrated superior performance in feature extraction over the UNet, with a mean Dice score and mean Jaccard score of 0.98 and 0.97, respectively, and standard deviations of 0.04 and 0.06 for benign tumors and 0.99 and 0.98 with standard deviations of 0.01 for malignant tumors. For the biomarker data, the combination of five machine learning models-KNN, logistic regression, SVM, decision tree, and random forest-resulted in an improved accuracy of 92.8% compared to single classifiers.

A novel framework for identifying causes of forest fire events using environmental and temporal characteristics of the ignition point in fire footprint

BackgroundForest fire is a very common natural or human-made disaster, which not only burns precious forest resources, but also further alters the forest ecosystem structure and functions as well as global climate. Accordingly, effective information extraction, data management, and modeling of historical fire events are imperative for forecasting and mitigating the onset and progression of forest fires. Traditional remote sensing-based forest fire detection and risk prediction operate at the pixel scale, overlooking the integrity of fire behaviour. Moreover, fire cause records typically stem from ground reports, not fully leveraging the advantages of high temporal resolution thermal infrared remote sensing products. MethodsWe proposed a framework to identify the causes of forest fire events, experimenting with fire incidents from 2001 to 2020 in the Daxing'anling region of Northeast China. Initially, the Jenk-DBSCAN model was employed to acquire fire footprints representative of forest fire events from MODIS MCD64A1 product, and footprint-based fire event attribute information such as ignition locations and timings. Fire footprints matching ground surveys were subsequently filtered, and corresponding environmental variable datasets (including meteorological, topographical, fuel, and human activity factors) were derived in consideration of the spatio-temporal differences of footprint origin and center points. Finally, the optimal variable combinations of the two datasets were fed to three binary classification models including Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM) respectively to identify natural fires or human-induced fires, followed by assessing the accuracy of the fire cause identification models. ResultsResults revealed that there was a high spatio-temporal consistency of those matched fire footprints between the model-extracted and the ground-recorded fire events, evidenced by an overall R2 of 0.87 derived from their area validation. The ignition dates extracted from fire footprints often preceded those reported by local agencies. Among all the selected variable combinations for fire cause identification, the meteorological variables dominated, with daily mean dew point (DEWP) and daily mean visibility (VISIB) being preferentially selected by most models. Across the six models, the model training AUC values were almost greater than 0.8, with the LR model demonstrating superior validation accuracy over RF and SVM. Notably, the accuracy of the LR model using the origin point based dataset surpassed that of the center point-based dataset. The best model achieved an overall accuracy of 90.48%, with the user’s accuracies for human-made fire and natural fire being at 92.86% and 85.71%, respectively. ConclusionsThe proposed framework offers a reliable method for enriching attribute information in forest fire management databases. The resulting fire footprints and origin points can assist in better characterizing the progression of forest fire events. Such endeavours help devise targeted forest fire prevention regulations, fostering sustainable forest management.

Residual facial erythema in atopic dermatitis patients treated with dupilumab stratified by machine learning.

Persistent facial erythema represents a significant complication in atopic dermatitis (AD) patients undergoing treatment with dupilumab. Stratifying patients based on the erythema course is crucial for elucidating heterogeneous phenotypes and facilitating advanced drug efficacy predictions. This study aimed to identify factors associated with facial erythema severity in dupilumab-treated AD patients and to establish a prediction model for drug response based on the identified factors. Data from a retrospective study conducted between July 2018 and July 2021 were collected and analysed. Patients were categorized into three groups via hierarchical clustering based on the course of facial erythema: early remission, low remission and persistent residual. LightGBM, a supervised gradient boosting decision tree algorithm, was employed to discern group differences and construct a prediction model. The model incorporated patient demographic and clinical profiles, including pre- and post-treatment examinations. The model's performance was evaluated using accuracy and the area under the receiver operating characteristic curve (AUC). The binary classification model demonstrated an accuracy of 89.10% and an AUC of 0.862 when distinguishing between early remission and persistent residual patients. The eight prominent factors associated with facial erythema severity included age, sex, lactate dehydrogenase (LDH), immunoglobulin E (IgE), eosinophil count, white blood cell count, Alnus allergy and cedar allergy. This study has two main significances: first, three clusters were identified through unsupervised learning; second, a classification model was constructed that proved more accurate than random prediction. The stratification and identification of crucial factors associated with residual facial erythema in dupilumab-treated AD patients lay the foundation for AI-powered prognostic models. This groundwork provides a substantial basis for enhancing future medical AI support in AD treatment selection, potentially improving personalized treatment approaches and outcomes.

Trajectory Interception Classification for Prediction of Collision Scope between Moving Objects

In the fields of autonomous navigation and vehicle safety, accurately predicting potential collision field points between moving objects is a significant challenge. A novel computing technique to enhance trajectory interception analysis is presented in this paper. Our objective is to develop a field model that can accurately forecast collision zones, improving road transportation safety and the use of autonomous cars. Our main contribution is a binary classification model called PCSMO (Prediction of Collision Scope between Moving Objects), which is based on zero-shot learning. Gann angles, which are typically 45 degrees, are used to analyze the trajectories of moving objects. This method is inspired by GANN (Gann Angle Numeric Nomenclature). Compared to earlier techniques, this model more accurately identifies potential collision collision interception zones. The technique computes Gann angles for trajectory analysis and extracts GPS coordinates of moving objects from video data using OpenCV. It offers a more sophisticated comprehension of object movement patterns and points of interception. To assess the precision, recall, F1-score, and prediction accuracy of our model, we employ 10-fold cross-validation. Comparing the PCSMO model to existing models, these metrics demonstrate how well the PCSMO model predicts potential collision zones. Our approach, we discovered, enhances trajectory analysis—a critical component of safer autonomous navigation systems. With potential applications in autonomous vehicle and UAV safety, the PCSMO model improves field interception classification.

Diabetic Retinopathy Detection Using InceptionResnet-V2 and Densenet121

This project addresses the global health challenge posed by the prevalence of diabetic retinopathy (DR) by developing an efficient automated diagnostic system. The dataset, consisting of diverse high-resolution retinal images, underwent preprocessing to categorize images into No DR (0) and DR (1-4) classes. The First initial binary classification model using a Convolutional Neural Network (CNN) discriminated between healthy and diseased retinas. Subsequently, The second multi-class CNN model was designed to predict the severity of diabetic retinopathy (DR) across a spectrum from mild (1) to proliferative DR (4), enabling a fine-grained analysis for early identification of cases requiring urgent intervention. To address real-world complexities, potential noise in the dataset, including artifacts and exposure variations, was acknowledged. The CNN models were designed to exhibit resilience to these challenges, ensuring robust performance in clinical settings. Preprocessing is considered the common occurrence of image inversion in retinal imaging by incorporating anatomical features, such as macula position and notches, to correctly identify image orientation and enhance result interpretability. The proposed automated analysis system demonstrated promising results in accurately categorizing retinal images into No DR and DR, as well as assigning severity scores for diabetic retinopathy. This project contributes significantly to computer-aided diagnostics, Supplying a dependable instrument for promptly identifying and addressing cases of diabetic retinopathy.

A sequence-based model for identifying proteins undergoing liquid-liquid phase separation/forming fibril aggregates via machine learning.

Liquid-liquid phase separation (LLPS) and the solid aggregate (also referred to as amyloid aggregates) formation of proteins, have gained significant attention in recent years due to their associations with various physiological and pathological processes in living organisms. The systematic investigation of the differences and connections between proteins undergoing LLPS and those forming amyloid fibrils at the sequence level has not yet been explored. In this research, we aim to address this gap by comparing the two types of proteins across 36 features using collected data available currently. The statistical comparison results indicate that, 24 of the selected 36 features exhibit significant difference between the two protein groups. A LLPS-Fibrils binary classification model built on these 24 features using random forest reveals that the fraction of intrinsically disordered residues (FIDR ) is identified as the most crucial feature. While, in the further three-class LLPS-Fibrils-Background classification model built on the same screened features, the composition of cysteine and that of leucine show more significant contributions than others. Through feature ablation analysis, we finally constructed a model FLFB (Feature-based LLPS-Fibrils-Background protein predictor) using six refined features, with an average area under the receiver operating characteristics of 0.83. This work indicates using sequence features and a machine learning model, proteins undergoing LLPS or forming amyloid fibrils can be identified.

In silico prediction of ocular toxicity of compounds using explainable machine learning and deep learning approaches.

The accurate identification of chemicals with ocular toxicity is of paramount importance in health hazard assessment. In contemporary chemical toxicology, there is a growing emphasis on refining, reducing, and replacing animal testing in safety evaluations. Therefore, the development of robust computational tools is crucial for regulatory applications. The performance of predictive models is heavily reliant on the quality and quantity of data. In this investigation, we amalgamated the most extensive dataset (4901 compounds) sourced from governmental GHS-compliant databases and literature to develop binary classification models of chemical ocular toxicity. We employed 12 molecular representations in conjunction with six machine learning algorithms and two deep learning algorithms to create a series of binary classification models. The findings indicated that the deep learning method GCN outperformed the machine learning models in cross-validation, achieving an impressive AUC of 0.915. However, the top-performing machine learning model (RF-Descriptor) demonstrated excellent performance with an AUC of 0.869 on the test set and was therefore selected as the best model. To enhance model interpretability, we conducted the SHAP method and attention weights analysis. The two approaches offered visual depictions of the relevance of key descriptors and substructures in predicting ocular toxicity of chemicals. Thus, we successfully struck a delicate balance between data quality and model interpretability, rendering our model valuable for predicting and comprehending potential ocular-toxic compounds in the early stages of drug discovery.

MisDetect: Iterative Mislabel Detection using Early Loss

Supervised machine learning (ML) models trained on data with mislabeled instances often produce inaccurate results due to label errors. Traditional methods of detecting mislabeled instances rely on data proximity, where an instance is considered mislabeled if its label is inconsistent with its neighbors. However, it often performs poorly, because an instance does not always share the same label with its neighbors. ML-based methods instead utilize trained models to differentiate between mislabeled and clean instances. However, these methods struggle to achieve high accuracy, since the models may have already overfitted mislabeled instances. In this paper, we propose a novel framework, MisDetect, that detects mislabeled instances during model training. MisDetect leverages the early loss observation to iteratively identify and remove mislabeled instances. In this process, influence-based verification is applied to enhance the detection accuracy. Moreover, MisDetect automatically determines when the early loss is no longer effective in detecting mislabels such that the iterative detection process should terminate. Finally, for the training instances that MisDetect is still not certain about whether they are mislabeled or not, MisDetect automatically produces some pseudo labels to learn a binary classification model and leverages the generalization ability of the machine learning model to determine their status. Our experiments on 15 datasets show that MisDetect outperforms 10 baseline methods, demonstrating its effectiveness in detecting mislabeled instances.

Thorough Assessment of Machine Learning Techniques for Predicting Protein-Nucleic Acid Binding Hot Spots

Background: Proteins and nucleic acids are vital biomolecules that contribute significantly to biological life. The precise and efficient identification of hot spots at protein-nucleic acid interfaces is crucial for guiding drug development, advancing protein engineering, and exploring the underlying molecular recognition mechanisms. As experimental methods like alanine scanning mutagenesis prove to be time-consuming and expensive, a growing number of machine learning techniques are being employed to predict hot spots. However, the existing approach is distinguished by a lack of uniform standards, a scarcity of data, and a wide range of attributes. Currently, there is no comprehensive overview or evaluation of this field. As a result, providing a full overview and review is extremely helpful. Methods: In this study, we present an overview of cutting-edge machine learning approaches utilized for hot spot prediction in protein-nucleic acid complexes. Additionally, we outline the feature categories currently in use, derived from relevant biological data sources, and assess conventional feature selection methods based on 600 extracted features. Simultaneously, we create two new benchmark datasets, PDHS87 and PRHS48, and develop distinct binary classification models based on these datasets to evaluate the advantages and disadvantages of various machine-learning techniques. Results: Prediction of protein-nucleic acid interaction hotspots is a challenging task. The study demonstrates that structural neighborhood features play a crucial role in identifying hot spots. The prediction performance can be improved by choosing effective feature selection methods and machine learning methods. Among the existing prediction methods, XGBPRH has the best performance. Conclusion: It is crucial to continue studying hot spot theories, discover new and effective features, add accurate experimental data, and utilize DNA/RNA information. Semi-supervised learning, transfer learning, and ensemble learning can optimize predictive ability. Combining computational docking with machine learning methods can potentially further improve predictive performance.

Prediction of operating state of hydrocyclones using vibrometry and 1D convolutional neural networks

Due to high processing capacity requirements and constraints on industrial water utilization, operation of hydrocyclones at higher feed solid concentration is often unavoidable. Hydrocyclone operation at such higher feed solid concentrations is accompanied by the risk of onset of rope discharge, adversely affecting both the size separation and thickening performance of the hydrocyclone. Numerous condition monitoring efforts have been reported in the literature regarding the development of online techniques for detection of roping in hydrocyclones. However, detection of roping, post its commencement, only ever allows for corrective maintenance in the form of measures taken for shifting the discharge back to the desired spray profile. This study focuses on the characterization and online detection of a transition discharge state which precedes the onset of roping, thereby making the prevention of onset of roping, possible. The transition discharge state was delineated from the conventional spray and rope discharge types based upon the observed differences in the decay rates of underflow discharge spray angles with increasing feed solid concentrations. A binary classification model (spray – rope) and a ternary classification model (spray – transition – rope) were developed with a 1D Convolutional neural network architecture capable of taking 5 secs of tri-axial (3-channel) vibration signal data as input, and outputting the predicted discharge type labels with an accuracy of 97.31 % and 94.94 %, respectively.

An Interactive Spreadsheet Model for Teaching Classification Using Logistic Regression

We present an interactive spreadsheet that supports teaching essential concepts in classification using the logistic regression (LoR) model for binary classification. The interactive spreadsheet demonstrates the capabilities of LoR by integrating computation with visualization. Students will reinforce concepts like probabilities, maximum likelihood estimation (MLE), and the use of likelihoods to optimize parameters for the LoR. We then discuss using LoR for classifications while adjusting its decision boundary (DB), demonstrating how to convert assigned likelihoods into classification using the DB; impact classification outcome by varying DBs; designate predictions as true positive, true negative, false positive, or false negative; and determine the classification accuracy. We use a variety of performance measures, including sensitivity, specificity, precision, negative predictive value, F1 and F2 scores, the receiver operating characteristics curve, and lift/decile charts. These measures are dynamically adjusted when the DB changes. We also reiterate the usage of these measures in the context of crossvalidation and imbalanced data sets. We provide a case study that implements LoR and an option for teaching the details behind MLE. We discuss the pedagogical aspects of this spreadsheet based on a survey of the 2022 student cohort in the Master of Management Analytics Program at the Rotman School of Management.