Binary Classification Techniques Research Articles

Bagging Vs. Boosting in Ensemble Machine Learning? An Integrated Application to Fraud Risk Analysis in the Insurance Sector

ABSTRACT Addressing the pressing challenge of insurance fraud, which significantly impacts financial losses and trust within the insurance industry, this study introduces an innovative automated detection system utilizing ensemble machine learning (EML) algorithms. The approach encompasses four strategic phases: 1) Tackling data imbalance through diverse re-sampling methods (Over-sampling, Under-sampling, and Hybrid); 2) Optimizing feature selection (Filtering, Wrapping, and Embedding) to enhance model accuracy; 3) employing binary classification techniques (Bagging and Boosting) for effective fraud identification; and 4) applying explanatory model analysis (Shapley Additive Explanations, Break-down plot, and variable-importance Measure) to evaluate the influence of individual features on model performance. Our comprehensive analysis reveals that while not every re-sampling technique improves model performance, all feature selection methods markedly bolster predictive accuracy. Notably, the combination of the Gradient Boosting Machine (GBM) algorithm with NCR re-sampling and GBMVI feature selection emerges as the most effective configuration, offering superior fraud detection capabilities. This study not only advances the theoretical framework for combating insurance fraud through AI but also provides a practical blueprint for insurance companies aiming to incorporate advanced AI strategies into their fraud detection arsenals, thereby mitigating financial risks and fostering trust systems.

Analisis Sentimen Ulasan Pada Aplikasi E-Commerce Shopee Dengan Menggunakan Algoritma Naïve Bayes

Every year, the e-commerce industry in Indonesia grows and develops at a rapid pace. In Indonesia, many online marketplaces have sprung up, including Tokopedia, Lazada, and Shopee. People are very interested in all e-commerce companies because they are hassle-free and instant. Among the most famous is Shopee, which offers a wide range of services and also presents a ratings and reviews column. This feature allows users to express their feelings about Shopee. Based on the information gathered from previous user comments, consumers can use these ratings to identify and trust both excellent and negative recommendations of the app they want to use. Sentiment analysis results include both favorable and negative user reviews by scoring, classifying, and filtering viewpoints to help businesses or users. The author uses the naïve bayes algorithm in this research. The Naïve Bayes Classifier approach will be used in this research to perform sentiment classification. The author then uses associations between frequently discussed word terms or themes that are related to each other for the extraction and exploration process, as well as descriptive statistics. Naïve Bayes Classifier is a binary classification technique that applies Bayesian principles with a strong assumption of independence, utilizing simple statistical probabilities.

Machine learning‐assisted study of correlation between post‐transition‐state bifurcation and initial phase information at the ambimodal transition state

AbstractThe Diels–Alder cycloaddition of cyclopentadiene and nitroethene, the intramolecular cycloaddition between a diene and triene, and the Diels–Alder cycloaddition of 2‐hydroxyacrolein with 1,3‐butadiene involving post‐transition‐state bifurcation (PTSB) were studied. These cycloaddition reactions were investigated using quasi‐classical trajectory (QCT), classical molecular dynamics (MD), ring‐polymer molecular dynamics (RPMD) simulations, and supervised machine‐learning binary classification techniques. Room‐temperature dynamics simulations started from the ambimodal transition state (TS) using the QCT, classical MD, and RPMD methods presented similar dynamics. Binary classification revealed that the initial geometry displacement from the ambimodal TS for the Diels–Alder cycloaddition of cyclopentadiene and nitroethene contributed to the branching dynamics and that the initial momenta for the intramolecular cycloaddition between a diene and triene and the Diels–Alder cycloaddition of 2‐hydroxyacrolein with 1,3‐butadiene played a significant role in the bifurcation dynamics.

Comparative analysis of binary and one-class classification techniques for credit card fraud data

The yearly increase in incidents of credit card fraud can be attributed to the rapid growth of e-commerce. To address this issue, effective fraud detection methods are essential. Our research focuses on the Credit Card Fraud Detection Dataset, which is a widely used dataset that contains real-world transaction data and is characterized by high class imbalance. This dataset has the potential to serve as a benchmark for credit card fraud detection. Our work evaluates the effectiveness of two supervised learning classification techniques, binary classification and one-class classification, for credit card fraud detection. The performance of five binary-class classification (BCC) learners and three one-class classification (OCC) learners is evaluated. The metrics used are area under the precision-recall curve (AUPRC) and area under the receiver operating characteristic curve (AUC). Our results indicate that binary classification is a better approach for detecting credit card fraud than one-class classification, with the top binary classifier being CatBoost.

Some Insights about the Applicability of Logistic Factorisation Machines in Banking

Logistic regression is a very popular binary classification technique in many industries, particularly in the financial service industry. It has been used to build credit scorecards, estimate the probability of default or churn, identify the next best product in marketing, and many more applications. The machine learning literature has recently introduced several alternative techniques, such as deep learning neural networks, random forests, and factorisation machines. While neural networks and random forests form part of the practitioner’s model-building toolkit, factorisation machines are seldom used. In this paper, we investigate the applicability of factorisation machines to some binary classification problems in banking. To stimulate the practical application of factorisation machines, we implement the fitting routines, based on logit loss and maximum likelihood, on commercially available software that is widely used by banks and other large financial services companies. Logit loss is usually used by the machine learning fraternity while maximum likelihood is popular in statistics. Depending on the coding of the target variable, we will show that these methods yield identical parameter estimates. Often, banks are confronted with predicting events that occur with low probability. To deal with this phenomenon, we introduce weights in the above-mentioned loss functions. The accuracy of our fitting algorithms is then studied by means of a simulation study and compared with logistic regression. The separation and prediction performance of factorisation machines are then compared to logistic regression and random forests by means of three case studies covering a recommender system, credit card fraud, and a credit scoring application. We conclude that logistic factorisation machines are worthy competitors of logistic regression in most applications, but with clear advantages in recommender systems applications where the number of predictors typically outnumbers the number of observations.

Investigation of Single and Multiple Mutations Prediction Using Binary Classification Approach

The mutation is a critical element in determining the proteins’ stability, becoming a core element in portraying the effects of a drug in the pharmaceutical industry. Doing wet laboratory tests to provide a better perspective on protein mutations is expensive and time-intensive since there are so many potential mutations, computational approaches that can reliably anticipate the consequences of amino acid mutations are critical. This work presents a robust methodology to analyze and identify the effects of mutation on a single protein structure. Initially, the context in a collection of words is determined using a knowledge graph for feature selection purposes. The proposed prediction is based on an easier and simpler logistic regression inferred binary classification technique. This approach can able to obtain a classification accuracy (AUC) Area Under the Curve of 87% when randomly validated against experimental energy changes. Moreover, for each cross-fold validation, the precision, recall, and F-Score are presented. These results support the validity of our strategy since it performs the vast majority of prior studies in this domain.

Characterizing and Detecting Methods to be Benchmarked under Performance Unit Test

Continuous integration is a growing trend in the software engineering community and industry. Performance testing is becoming more important in this context. To support precise and fine-grained monitoring, performance unit tests are applied for small software components. However, the benchmarks for performance unit testing are still insufficient, which means that benchmark coverage is low and there is a room for improvement. Therefore, focusing on the most important parts of the software, such as methods, and ensuring that their performance is monitored closely with performance unit tests can greatly reduce the amount of work that needs to be done for testing and to prepare benchmarks. This paper aims to provide an assisting approach for detecting methods that need to be benchmarked in performance unit tests. We start by defining 30 features to characterize the methods in the projects and show that they can be used to tell the benchmarked methods (short for BDMs) from those that are not. Then, using the proposed features, we build machine learning-based models to detect BDMs. We perform an experiment with 10 open source projects from GitHub to see how well our approach works. First, we use seven binary classification techniques to evaluate the prediction performance of our machine learning models. We find that Random Forest makes the best predictions where AUC and MCC are between 0.77 and 0.89 and 0.5 and 0.75, respectively. In terms of cost effectiveness, the experiment reveals that by inspecting only 5% of the candidate methods detected by our model, 43% of the total real BDMs can be retrieved. Second, we conduct feature importance evaluations for individual features and feature categories. We find that eight features related to Scope, History, and Complexity are individually important for good predictions and that the combination of all features in the Scope category is paramount for our model, while the combination of features in the Control Flow category is less important. Third, we investigate the performance of our detection approach with different feature selection strategies and data sources. Our results show that we can make good predictions about whether a method needs to be benchmarked by using machine learning models. Practitioners can use our method and the results of the study to deal with BDMs detection effectively.

IoT-based multimodal liveness detection using the fusion of ECG and fingerprint

Purpose Biometric scans using fingerprints are widely used for security purposes. Eventually, for authentication purposes, fingerprint scans are not very reliable because they can be faked by obtaining a sample of the fingerprint of the person. There are a few spoof detection techniques available to reduce the incidence of spoofing of the biometric system. Among them, the most commonly used is the binary classification technique that detects real or fake fingerprints based on the fingerprint samples provided during training. However, this technique fails when it is provided with samples formed using other spoofing techniques that are different from the spoofing techniques covered in the training samples. This paper aims to improve the liveness detection accuracy by fusing electrocardiogram (ECG) and fingerprint. Design/methodology/approach In this paper, to avoid this limitation, an efficient liveness detection algorithm is developed using the fusion of ECG signals captured from the fingertips and fingerprint data in Internet of Things (IoT) environment. The ECG signal will ensure the detection of real fingerprint samples from fake ones. Findings Single model fingerprint methods have some disadvantages, such as noisy data and position of the fingerprint. To overcome this, fusion of both ECG and fingerprint is done so that the combined data improves the detection accuracy. Originality/value System security is improved in this approach, and the fingerprint recognition rate is also improved. IoT-based approach is used in this work to reduce the computation burden of data processing systems.

Development of a three tiered cognitive hybrid machine learning algorithm for effective diagnosis of Alzheimer’s disease

Alzheimer's disease (AD) is one of the most frequent neurodegenerative disorders in the elderly subjects. Since early detection can prevent or delay cognitive decline in the older subjects, it is desirable to develop effectual protocols for the diagnosis of the disease. Most of the existing diagnostic tools fail to improvise timely disease prognosis in susceptible patients. Keeping this fact into consideration, we developed a cognitive-based 3-tiered machine learning (ML) algorithm employing baseline characteristics to predict AD or mild cognitive impairment (MCI) to construct psychometric test results. Earlier machine learning based AD diagnosis methods used a binary or multinomial classification technique. We relied on the development of a sophisticated hybrid cognitive ML algorithm that provides an accurate and precise prediction of the disease. We built an ML model using cognitive and demographic data. The prediction method consisted of a three-step process. Alzheimer’s Disease Neuroimaging Initiative (ADNI) database was used to develop a novel prediction algorithm. Considering the fact that nineteen ML and deep learning classifiers could not adequately classify ADNI data, we created a 2-layer model stacking procedure. Model stacking outperformed six ML classifier combinations, including Logistic Regression, Naïve Bayes, Support Vector Machine, Decision Trees, Random Forest, and eXtreme Gradient Boosting. The performance of the as-proposed model was evaluated employing seven performance assessment measures and four classification error indicators. Each model was evaluated in three separate strategical assessment modules. In the first experiment, XGB, Random Forest, and SVM achieved 89.63% accuracy, while Random Forest achieved 93.90% accuracy in the second experiment. Experiment 2 improved the classification and performance of overall prediction. In the third experiment, hybrid modeling, the accuracy increased significantly, with experiment 1 giving 90.24% accuracy and experiment 2 yielding 95.12% accuracy. The as-proposed model successfully predicted early AD and MCI in an effective manner. We were able to reduce nineteen classifiers into four classifiers (from experiment-1) and six classifiers (from experiment-2) and subsequently into one meta-learner (19 → 4 → 1 and 19 → 6 → 1), with high predictive power. Finally, we performed a thorough comparative analysis of different ADNI datasets to validate our findings.

A Deep Learning-Based Compact Weighted Binary Classification Technique to Discriminate between Targets and Clutter in SAR Images

The proposed approach is a deep learning-based compact weighted binary classification (DL-CWBC) method to discriminate between targets and clutter in synthetic aperture radar (SAR) images. A new modified cross-entropy error function is proposed to improve the probability of detection by controlling the rate of false alarms (FAs). The unique feature of a CWBC algorithm is reducing the FA rate and maximizing the probability of target detection without missing any target. For pre-processing, targets and clutter are detected through a constant false alarm rate (CFAR) as a conventional detection algorithm. These are then manually divided into two classes. The classified targets and clutter were trained through a ResNet-101 network. There is a trade-off between the minimization of the FA rate and the maximization of the detection probability for targets of interest (TOIs). The weighted coefficient of the modified cross-entropy error function tries to maximize the performance of this trade-off. In addition, the proposed approach enables us not to miss any targets by an extreme distinction decision. Above all, the DL-CWBC algorithm performs very well despite its simplicity.

Predictive Analytics Using Machine Learning to Identify ART Clients at Health System Level at Greatest Risk of Treatment Interruption in Mozambique and Nigeria.

A core objective of HIV/AIDS programming is keeping clients on treatment to improve their health outcomes and to limit spread. Machine learning and artificial intelligence can combine client, temporal, and locational attributes to identify which clients are at greatest risk of loss to follow-up (LTFU) and enable health providers to direct support interventions accordingly. The analysis was part of a project funded by U.S. President's Emergency Plan for AIDS Relief and United States Agency for International Development, Data for Implementation, and applied to data from publicly available sources (health facility data, geospatial data, and satellite imagery) and de-identified electronic medical record data on antiretroviral therapy clients in Nigeria and Mozambique. The project applied binary classification techniques using temporal cross-validation to predict the risk that patients would be LTFU. Classifiers included logistic regression, neural networks, and tree-based models. Models showed strong predictive power in both settings. In Mozambique, the best-performing model, a Random Forest, achieved an area under the precision-recall curve of 0.65 compared against an underlying LTFU rate of 23%. In Nigeria, the best-performing model, a boosted tree, achieved an area under the precision-recall curve of 0.52 compared against an underlying LTFU rate of 27%. Machine-learned models outperformed current classification techniques and showed potential to better direct health worker resources toward patients at greatest risk of LTFU. Moreover, models performed equally across sex and age groups, supporting the model's generalizability and wider application.

Automation and smart materials in detecting smart contracts vulnerabilities in Blockchain using deep learning

Smart contracts are a vital component of applications being built by blockchain or distributed ledger technology. Smart contracts consists of computer code that composed set of rules agreed upon by the involved parties. When these predefined conditions or rules are satisfied by the transactions initiated by parties, the smart contract executes itself to complete the transaction. Normally, while performing transactions with agreements involving multiple parties, a third party have to be involve to verify all the information, which makes it complex and time consuming process. Smart contracts simplify this process by eliminating the third party and automating the process, enabling the stakeholders to perform transactions directly with each other. Smart contract itself is replicated among multiple nodes of a blockchain there by giving benefits of immutability, security and permanence. Most smart contracts are written in Solidity programming language due to its simplicity and conciseness. Attackers are attracted by the popularity of Solidity language and its vulnerability possibilities. For example, in the 2016 year, 60 million US dollars was theft by Decentralized Autonomous Organization (DAO) attack due to vulnerabilities present in smart contracts. There are few existing tools and papers on this area to find vulnerabilities on smart contracts but taking more time to predict. Thus, this paper presents different Deep learning techniques with satisfactory results to identify smart contract vulnerabilities which are Re-entrancy, Denial of Service (DOS) and Transaction Origin using binary, multi class and multi label classification techniques.

Detecting and filtering rumor in social media using news media event

SummaryIn social media platform, many users post messages to express their interests and preferences daily. Because of its fast and easy access abundant people follow news events and there is a possibility for spreading rumor or fake news. This fake news is unverified at the time of posting. Therefore, it is necessary to detect and remove the fake news before it spread widely. Rumors or fake news are created illegally for the purpose of popularity, hike in their business or financial, and so forth, and these rumors need to be detected as easily as possible. Our proposed rumor detection method compares the social media content with news media and applies the support vector machine (SVM) as binary classification technique. Our experiments results revealed that the proposed method attained considerable improvement when compared to the existing machine learning techniques. The proposed SVM rumor detection approach attained better results of 89% precision, 64% recall, and 85% F‐measure. The experimental results proved that the proposed method will be useful for avoiding social damages caused by rumors in social media.

Adjusted binary classification (ABC) model in forensic science: An example on sex classification from handprint dimensions

Binary classification techniques are commonly used in forensic examination to test if a specimen belongs to a particular group and base the expert opinion on the questioned evidence. However, most of the currently used methods do not achieve sufficient accuracy due to the ignoring of the specimens classified in the overlapping area. To address the issue, we proposed a novel Adjusted binary classification (ABC) algorithm that automatically adjusts posterior probabilities to reach classification accuracy and positive/negative predicted values (PPV, NPV) of 95 %.In the presented example, we used three handprint measurements from 160 participants (80 males and 80 females) to develop models that would classify sex from their dimensions. The sample was split into the training/cross-validated (70 %) and testing sample (30 %). We developed four classification models using linear discriminant analysis (LDA) by employing traditional single cut-off values and ABC approach that for each group provides a specific posterior probability cut-off threshold.In the cross-validated sample, the accuracy of traditional models was 78.7–92.5 %, while PPVs/NPVs ranged between 78.2 and 93 %. ABC models provided 95 % accuracy, PPV, and NPV, and could classify 35.5–88.1 % of specimens. In the testing sample, ABC models achieved accuracy of 97.3–100 %, PPV/NPV 95.4–100 %, and could be applied to 29.1–87.5 % of specimens.The study demonstrated that the ABC approach could adjust classification models to reach predefined values of accuracy, PPV, and NPV. Therefore, it could be an efficient tool for conducting a binary classification in forensic settings and minimizing the possibilities of incorrect classifications.

Feature analysis for data-driven APT-related malware discrimination

Advanced Persistent Threats (APTs) have become a major concern for IT security professionals around the world. These attacks are characterized by the use of both highly sophisticated, evasive and cautious human and technical resources. It is very common to notice the combined use of different malware in long APT campaigns. This fact makes it interesting to investigate the malware that has been used in APT campaigns. Different approaches have been proposed to find discriminatory features to detect APT malware. Features from either static, dynamic and network-related analyses have been separately proposed for that aim. The new approach considered in this study aims to identify the most discriminatory features to distinguish APT-campaign-belonging malware from non-APT malware executables. This approach suggests to identify the discriminatory features from not one but all three groups of these analyses by using domain knowledge and with a purpose of interpretability. As a result, a set with the most discriminatory features of each type is provided. To achieve this set, well-known machine learning techniques have been used. One of the most important limitations in the use of these learning techniques is the availability of a relevant amount of data. In this paper, a large dataset of 19,457 malware samples is publicly provided, including both malware known to be related with APTs and generic non-APT-belonging malware samples. In order to analyze the discrimination ability of the features, the proposed approach follows several steps. First, an exploratory analysis is conducted to obtain knowledge about the data structure. Later, feature selection is performed using different discriminatory techniques. The resulting selection of features is assessed by means of four well-known binary classification techniques. The high accuracy of the results shows that the proposed features are discriminative enough for the stated purpose. Finally, these results are interpreted and the findings are discussed from the perspective of prior knowledge and assumptions about APT-related malware.

Classification of rainfall data using support vector machine

Classification problems can be based on cross-section or time-series data. In general, some of the characteristics found in time series data are data that are very susceptible to containing noise and outliers. In this paper, time series data-based classification was carried out using the support vector machine (SVM) method. The SVM method is the most popular binary classification technique in machine learning. The advantage of this method is that it can find a global optimum solution and always achieve the same solution for every running. Another advantage is that it can solve the over-fitting problem by minimizing the upper limit of generalization errors. SVM classification model performance can be seen from classification accuracy and sensitivity and specificity tests. The results showed that based on the level of classification accuracy used the SVM method resulted in an accuracy rate of the prediction results with training data of 96,3% and the accuracy of the prediction results with testing data of 90,08%. Therefore, the classification of rainfall data using the SVM method has a very good performance.

Multi-Machine Learning Binary Classification, Feature Selection and Comparison Technique for Predicting Death Events Related to Heart Disease

Multi-Machine Learning Binary Classification, Feature Selection and Comparison Technique for Predicting Death Events Related to Heart Disease

A Machine Learning Approach for Ecg Analysis for Emotions

Emotions are feelings which one can feel and are hard to be put in a form by a person .However they reflect the mental state of a person. Emotions like joy and sadness can be somehow detected from the facial expressions or through the body language. But these emotions do have an impact upon our system. An individual’s electrocardiogram is a way through which one can know the impact of different parameters such as stress, joy, sadness, anger on the mechanism of our body. The emotions such as anger, sadness have an adverse effect on the cardio system and is seen in the form of abnormal ECG which can be a good pointer to a counselor when finding out the reasons and diagnosis. The decomposition technique along with the Hilbert transform can be used for feature retrieval. The different emotions are detected through the binary classification technique

A two-step machine learning approach to predict S&P 500 bubbles

ABSTRACT In this paper, we are interested in predicting the bubbles in the S&P 500 stock market with a two-step machine learning approach that employs a real-time bubble detection test and support vector machine (SVM). SVM as a nonparametric binary classification technique is already a widely used method in financial time series forecasting. In the literature, a bubble is often defined as a situation where the asset price exceeds its fundamental value. As one of the early warning signals, prediction of bubbles is vital for policymakers and regulators who are responsible to take preemptive measures against the future crises. Therefore, many attempts have been made to understand the main factors in bubble formation and to predict them in their earlier phases. Our analysis consists of two steps. The first step is to identify the bubbles in the S&P 500 index using a widely recognized right-tailed unit root test. Then, SVM is employed to predict the bubbles by macroeconomic indicators. Also, we compare SVM with different supervised learning algorithms by using k-fold cross-validation. The experimental results show that the proposed approach with high predictive power could be a favourable alternative in bubble prediction.

Profit-based churn prediction based on Minimax Probability Machines

In this paper, we propose three novel profit-driven strategies for churn prediction. Our proposals extend the ideas of the Minimax Probability Machine, a robust optimization approach for binary classification that maximizes sensitivity and specificity using a probabilistic setting. We adapt this method and other variants to maximize the profit of a retention campaign in the objective function, unlike most profit-based strategies that use profit metrics to choose between classifiers, and/or to define the optimal classification threshold given a probabilistic output. A first approach is developed as a learning machine that does not include a regularization term, and subsequently extended by including the LASSO and Tikhonov regularizers. Experiments on well-known churn prediction datasets show that our proposal leads to the largest profit in comparison with other binary classification techniques.