Fragility Index: A New Approach for Binary Classification

Understanding the reliability of outcomes in randomized controlled trials (RCTs) is crucial, as standard metrics, such as P- value do not fully capture result fragility. This led to the adoption of specific indices: the fragility index (FI), which measures the strength of trial through significant results by calculating the minimum number of patient status changes from nonevent to event required to make the results statistically insignificant; reverse fragility index (RFI), used for insignificant results; and continuous fragility index (CFI), which acts similar to FI for significant continuous outcomes. The objective is to assess the robustness of orthopaedics RCTs using these indices across leading orthopaedic journals. A systematic review of RCTs published between 2019 and 2023 in the top five general orthopaedic journals, identified through Scientific Journal Rankings, was done. Data extraction focused on FI, RFI, and CFI and related measures for 160 RCTs. The indices were calculated using established methodologies, with sample size adjustments. 22 RCTs had statistically significant dichotomous primary outcomes and 17 studies had notable dichotomous secondary outcomes. Twenty-nine had negligible (insignificant) dichotomous primary outcomes, and 92 reported notable continuous outcomes. Only one RCT reported a FI in the article. The median FI was 5 with a median sample size of 142 for dichotomous outcomes. The median RFI was 3 with a median sample size of 100 for negligible outcomes, and the median CFI was 13 with a median sample size of 86.5 for continuous outcomes, showing dichotomous outcomes to be more fragile than continuous ones. Continuous outcomes are less fragile than dichotomous outcomes, with negligible dichotomous outcomes being particularly more fragile. This fragility stems from small sample sizes and limited outcome events. Using these indices, especially when considering patient loss to follow-up, can improve the reliability of findings. I.

This article discusses the practical aspects of applying logistic regression for binary data classification. Logistic regression determines the probability of an object belonging to one of two classes. This probability is calculated with the help of a sigmoid function, the argument of which is a linear convolution of the feature vector of the object with the weighting coefficients obtained during the minimization of the logarithmic loss function. Predicted class labels are determined by comparing the calculated probability with a given threshold value. The logistic regression study was performed using the computer simulation method. For this, a software complex was developed, the work of which reproduces the main stages of logistic regression: preparation of input data, training, testing with determination of quality metrics of binary classification, application of the logistic regression method for data classification in practice. The paper examines the effect of overlapping and imbalance of classes in the input data set on the efficiency of binary classification. The overlapping of classes is modeled by the formation of input data based on two shifted relative to each other density functions of the normal distribution of random variables. Class imbalance is simulated by the probability of switching between these features. It is shown that when the distance between the mathematical expectations of the density functions of the normal distribution decreases or when the dispersion of random variables increases, the overlapping of relevant classes increases, which leads to an increase in the number of objects that the classifier can assign to one or another class. Approaching the probability of switching between the distribution functions of random variables to the extreme values of the unit interval leads to an increase in class imbalance, which is manifested in an increase in the number of elements of the input data set labeled with the label of the same class. It has been experimentally confirmed that the AUC ROC metric, popular in binary classification problems, is dependent on the degree of class overlap and relatively resistant to class imbalance.

Alzheimer's disease (AD) is a progressive neurodegenerative disorder that primarily affects older adults, leading to memory loss, cognitive decline, and behavioural changes.It is marked by the accumulation of amyloid-beta plaques and tau tangles in the brain, which impair neuronal function and eventually cause cell death.Diagnosis typically involves clinical evaluations, cognitive assessments, and imaging techniques such as MRI (Magnetic Resonance Imaging) and PET (Positron Emission Tomography) scans.Early detection is crucial, as it allows for interventions that can potentially slow the disease's progression, enabling patients and their families to manage symptoms more effectively.The transition from Normal Control (NC) to Early Mild Cognitive Impairment (EMCI) may be reversible, whereas other stages are not.This makes it a particularly challenging classification problem, as the changes in biomarker values are typically minimal during the progression from NC to very mild stages.To address this issue, we created a hybrid deep learning model, GDP (combining GoogLeNet, DenseNet-121, and Principal Component Analysis (PCA)), to extract hybrid features using 26639 sMRI images from well-known MRI datasets, such as the Open Access Series of Imaging Studies (OASIS) and the Alzheimer's Disease Neuroimaging Initiative (ADNI).These features fed into the SVM binary classification algorithm to distinguish very mild cases from healthy images.We created synthetic images using CycleGAN augmentation to balance the dataset across labels and expand its size.The proposed GDP model stands out from current state-of-the-art deep learning techniques, such as ResNet-50, DenseNet-169, VGG-16, EfficientNet, and Inceptionv3, by achieving an impressive accuracy of 98.98% and an AUC of 99.91%.These results significantly surpass the performance of existing methods, making it a promising tool for early Alzheimer's disease prediction.

Imaging spectroscopy as a tool to study sediment characteristics on a tidal sandbank in the Westerschelde

The UK financial sector increasingly employs machine learning techniques to enhance revenue and understand customer behaviour. In this study, we develop a machine learning workflow for high classification accuracy and improved prediction confidence using a binary classification approach on a publicly available dataset from a Portuguese financial institution as a proof of concept. Our methodology includes data analysis, transformation, training, and testing machine learning classifiers such as Naïve Bayes, Decision Trees, Random Forests, Support Vector Machines, Logistic Regression, Artificial Neural Networks, AdaBoost, and Gradient Descent. We use stratified k-folding (k=5) cross-validation and assemble the top-performing classifiers into a decision-making committee, resulting in over 92% accuracy with two-thirds voting confidence. The workflow is simple, adaptable, and suitable for UK banks, demonstrating the potential for practical implementation and data privacy. Future work will extend our approach to UK banks, reformulate the problem as a multi-class classification, and introduce pre-training automated steps for data analysis and transformation.

BackgroundTo harness the full potential of social media for epidemiological surveillance of drug abuse trends, the field needs a greater level of automation in processing and analyzing social media content.ObjectivesThe objective of the study is to describe the development of supervised machine-learning techniques for the eDrugTrends platform to automatically classify tweets by type/source of communication (personal, official/media, retail) and sentiment (positive, negative, neutral) expressed in cannabis- and synthetic cannabinoid–related tweets.MethodsTweets were collected using Twitter streaming Application Programming Interface and filtered through the eDrugTrends platform using keywords related to cannabis, marijuana edibles, marijuana concentrates, and synthetic cannabinoids. After creating coding rules and assessing intercoder reliability, a manually labeled data set (N=4000) was developed by coding several batches of randomly selected subsets of tweets extracted from the pool of 15,623,869 collected by eDrugTrends (May-November 2015). Out of 4000 tweets, 25% (1000/4000) were used to build source classifiers and 75% (3000/4000) were used for sentiment classifiers. Logistic Regression (LR), Naive Bayes (NB), and Support Vector Machines (SVM) were used to train the classifiers. Source classification (n=1000) tested Approach 1 that used short URLs, and Approach 2 where URLs were expanded and included into the bag-of-words analysis. For sentiment classification, Approach 1 used all tweets, regardless of their source/type (n=3000), while Approach 2 applied sentiment classification to personal communication tweets only (2633/3000, 88%). Multiclass and binary classification tasks were examined, and machine-learning sentiment classifier performance was compared with Valence Aware Dictionary for sEntiment Reasoning (VADER), a lexicon and rule-based method. The performance of each classifier was assessed using 5-fold cross validation that calculated average F-scores. One-tailed t test was used to determine if differences in F-scores were statistically significant.ResultsIn multiclass source classification, the use of expanded URLs did not contribute to significant improvement in classifier performance (0.7972 vs 0.8102 for SVM, P=.19). In binary classification, the identification of all source categories improved significantly when unshortened URLs were used, with personal communication tweets benefiting the most (0.8736 vs 0.8200, P<.001). In multiclass sentiment classification Approach 1, SVM (0.6723) performed similarly to NB (0.6683) and LR (0.6703). In Approach 2, SVM (0.7062) did not differ from NB (0.6980, P=.13) or LR (F=0.6931, P=.05), but it was over 40% more accurate than VADER (F=0.5030, P<.001). In multiclass task, improvements in sentiment classification (Approach 2 vs Approach 1) did not reach statistical significance (eg, SVM: 0.7062 vs 0.6723, P=.052). In binary sentiment classification (positive vs negative), Approach 2 (focus on personal communication tweets only) improved classification results, compared with Approach 1, for LR (0.8752 vs 0.8516, P=.04) and SVM (0.8800 vs 0.8557, P=.045).ConclusionsThe study provides an example of the use of supervised machine learning methods to categorize cannabis- and synthetic cannabinoid–related tweets with fairly high accuracy. Use of these content analysis tools along with geographic identification capabilities developed by the eDrugTrends platform will provide powerful methods for tracking regional changes in user opinions related to cannabis and synthetic cannabinoids use over time and across different regions.

Recent studies have successfully reported the accuracy of using artificial neural networks to predict grip force in controlled settings. However, only relying on accuracy to evaluate the machine learning models may lead to overoptimistic results, especially on imbalanced datasets. The Matthews correlation coefficient (MCC) showed an advantage in capturing all the data characteristics in the confusion matrix. Therefore, a binary classification approach and the MCC value were introduced to assess the performance of previously proposed machine learning models. Our results show that the overall correlations ranging between 0.48 and 0.59 indicate a strong relationship between predictions and actual scenarios. The binary classification approach and the MCC values could be used for future performance comparison with other machine learning models.

Schedae Informaticae » 2017 » Volume 26 » Failures prediction based on performance monitoring of a gas turbine: a binary classification approach A A A

The promise of machine learning successfully exploiting digital phenotyping data to forecast mental states in psychiatric populations could greatly improve clinical practice. Previous research focused on binary classification and continuous regression, disregarding the often ordinal nature of prediction targets derived from clinical rating scales. In addition, mental health ratings typically show important class imbalance or skewness that need to be accounted for when evaluating predictive performance. Besides it remains unclear which machine learning algorithm is best suited for forecast tasks, the eXtreme Gradient Boosting (XGBoost) and long short-term memory (LSTM) algorithms being 2popular choices in digital phenotyping studies. The CrossCheck dataset includes 6,364 mental state surveys using 4-point ordinal rating scales and 23,551 days of smartphone sensor data contributed by patients with schizophrenia. We trained 120 machine learning models to forecast 10 mental states (e.g., Calm, Depressed, Seeing things) from passive sensor data on 2 predictive tasks (ordinal regression, binary classification) with 2 learning algorithms (XGBoost, LSTM) over 3 forecast horizons (same day, next day, next week). A majority of ordinal regression and binary classification models performed significantly above baseline, with macro-averaged mean absolute error values between 1.19 and 0.77, and balanced accuracy between 58% and 73%, which corresponds to similar levels of performance when these metrics are scaled. Results also showed that metrics that do not account for imbalance (mean absolute error, accuracy) systematically overestimated performance, XGBoost models performed on par with or better than LSTM models, and a significant yet very small decrease in performance was observed as the forecast horizon expanded. In conclusion, when using performance metrics that properly account for class imbalance, ordinal forecast models demonstrated comparable performance to the prevalent binary classification approach without losing valuable clinical information from self-reports, thus providing richer and easier to interpret predictions.

Hyperspectral imaging (HSI) provides rich spectral information that is well-suited for human detection in complex environments such as disaster sites. However, the high dimensionality of HSI data imposes computational challenges. In this study, we propose a 3D convolutional neural network (3D CNN)-based approach for binary human presence classification using hyper-spectral data. To reduce the computational load while retaining key spectral features, we apply a band reduction technique based on spectral pooling within the biologically significant 500–700 nm wavelength range. Our model achieves 94% accuracy in controlled test environments, demonstrating strong potential for real-world applications in victim detection.

Steganalysis looks for concealed content in a digital file, which becomes more difficult to find using sophisticated embedding methods. Deep learning models especially convolutional neural networks (CNNs) have become rather effective methods for directly learning discriminative features from picture data without human involvement in recent years. Using the InceptionResNetV2 architecture a high capacity hybrid CNN model that combines the strengths of Inception modules (for multi scale feature extraction) and residual connections (to combat vanishing gradients and support deeper networks). We propose in this work a deep learning based steganalysis method. Using a labelled dataset of 7,267 RGB images resized to 150×150 pixels, our binary classification model separates between "stego" and "clean" images. Using pretrained ImageNet weights for the base network and training a bespoke classification head, the model was refined utilising transfer learning. With a total number of trainable parameters restricted to 196,994, the model attained an accuracy of 86.17%, a sensitivity of 96.95%, and an F1 score of 0.65 despite training for just 15 epochs, therefore demonstrating both efficacy and computational efficiency. The trade off is discussed with an eye towards architectural improvement and training technique development. The results confirm the feasibility of InceptionResNetV2 as a backbone for scalable, high accuracy steganalysis systems appropriate for real time picture forensics and cybersecurity uses.

Deep-discovery: Anomaly discovery in software-defined networks using artificial neural networks

Breast cancer is widely prevalent in women. To diagnose it, pathologists evaluate breast histopathology images under a microscope in various magnifications. However, the number of pathologists per population is low in many countries, and a mistake is probable. Nowadays, deep learning has become a popular artificial intelligence research trends. Even though deep neural networks have obtained promising results in image processing, the need for extensive training data avoids its use in medical image processing. After the evaluation of various state-of-the-art deep learning methods and algorithms in medical data processing, this study proposes an effective deep transfer learning-based model, relying on pre-trained DCNN using an extensive data of ImageNet dataset which improves state-of the-art systems in both binary and multi-class classification. The weights of the pre-trained DesneNet121 on the Imagenet are transferred as initial weights first, and then the model is fine-tuned with a deep classifier along with data augmentation to distinguish various malignant and benign samples in the two categories of binary and multi-class classification. In the multi-class classification, the proposed model obtained up to 97% image-level accuracy. In binary classification, the model obtained up to 100% image-level accuracy. The achieved-results are outperforming previous studies accuracies in multiple performance metrics in breast cancer CAD systems. Moreover, the proposed method is both flexible and scalable, meaning it can be easily expanded to cover the detection of other types of diseases in the future and be integrated with more CNNs to increase its generalization capabilities.

Introduction: We propose a novel approach for the assessment of the similarity of retinal vessel segmentation images that is based on linking the standard performance metrics of a segmentation algorithm, with the actual structural properties of the images through the fractal dimension.Method: We apply our methodology to compare the vascularity extracted by automatic segmentation against manually segmented images.Results: We demonstrate that the strong correlation between the standard metrics and fractal dimension is preserved regardless of the size of the subimages analyzed.Discussion or Conclusion: We show that the fractal dimension is correlated to the segmentation algorithm’s performance and therefore it can be used as a comparison metric.

Background/Objectives: Standardized reporting of astigmatism outcomes is essential for comparability, reproducibility, and interpretation of refractive surgery studies. Vectorial analyses based on established metrics are increasingly required by major journals, yet no freely available tool exists for generating publication-ready vector analysis tables with statistical comparisons. This study presents AstigMETRICS, a standalone application for automated calculation, formatting, and statistical comparison of standard vector metrics in refractive surgery. Methods: AstigMETRICS was developed in MATLAB and compiled as a standalone executable requiring no programming knowledge. The software accepts preoperative, intended, and postoperative astigmatism data in spreadsheet format for both refractive and corneal measurements. It calculates seven standard vector metrics following the Alpins method: the target-induced astigmatism (TIA), surgically induced astigmatism (SIA), difference vector (DV), correction index (CI), magnitude of error (ME), angle of error (AE), and index of success (IOS). Statistical comparisons are performed automatically using appropriate parametric or nonparametric tests for paired and unpaired study designs, with p-values and Cohen's d effect sizes reported. Results: AstigMETRICS generates standardized tables reporting the means, standard deviations, and clinically relevant proportions (percentage of eyes with an ME within ±0.50 D or ±1.00 D, and an AE within ±15°). Three simulated datasets were created to validate the software functionality across common surgical scenarios: a contralateral eye laser vision correction, toric phakic IOL implantation, and cataract surgery with toric IOLs. The output tables are displayed in standardized format and saved as high-resolution TIFF images suitable for publication. The software is freely available and a download link is provided in this article. Conclusions: AstigMETRICS enables clinicians and researchers to perform standardized, reproducible astigmatism vector analyses with built-in statistical comparisons. This freely available tool simplifies outcome reporting and improves methodological consistency in refractive surgery research.