Wrapper Method Research Articles

Benchmark study of feature selection strategies for multi-omics data

BackgroundIn the last few years, multi-omics data, that is, datasets containing different types of high-dimensional molecular variables for the same samples, have become increasingly available. To date, several comparison studies focused on feature selection methods for omics data, but to our knowledge, none compared these methods for the special case of multi-omics data. Given that these data have specific structures that differentiate them from single-omics data, it is unclear whether different feature selection strategies may be optimal for such data. In this paper, using 15 cancer multi-omics datasets we compared four filter methods, two embedded methods, and two wrapper methods with respect to their performance in the prediction of a binary outcome in several situations that may affect the prediction results. As classifiers, we used support vector machines and random forests. The methods were compared using repeated fivefold cross-validation. The accuracy, the AUC, and the Brier score served as performance metrics.ResultsThe results suggested that, first, the chosen number of selected features affects the predictive performance for many feature selection methods but not all. Second, whether the features were selected by data type or from all data types concurrently did not considerably affect the predictive performance, but for some methods, concurrent selection took more time. Third, regardless of which performance measure was considered, the feature selection methods mRMR, the permutation importance of random forests, and the Lasso tended to outperform the other considered methods. Here, mRMR and the permutation importance of random forests already delivered strong predictive performance when considering only a few selected features. Finally, the wrapper methods were computationally much more expensive than the filter and embedded methods.ConclusionsWe recommend the permutation importance of random forests and the filter method mRMR for feature selection using multi-omics data, where, however, mRMR is considerably more computationally costly.

Machine Learning Approach to Support the Detection of Parkinson's Disease in IMU-Based Gait Analysis

Recent advances in data analysis and wearable sensors for human movement monitoring can promote objective and accurate clinical evaluation of neurological symptoms and improve outcome measures in clinical trials [1–3]. The aim of this study was to combine modern technique of data analysis and wearable sensors to determine which supervised machine learning (ML) algorithm can most accurately classify people with Parkinson’s disease (pwPD) from speed-matched healthy subjects (HS) based on a selected minimum set of IMU-derived gait features. Twenty-two gait features were extrapolated from the trunk acceleration patterns of 81 pwPD and 80 HS, including spatiotemporal, pelvic kinematics, and acceleration-derived gait stability indexes. After a three-level feature selection procedure, seven gait features were considered for implementing five ML algorithms: support vector machine (SVM), artificial neural network, decision trees (DT), random forest (RF), and K-nearest neighbors. Accuracy, precision, recall, F1 score, AUC and generalization error were calculated. SVM outperformed the other ML algorithms in terms of classification metrics (test accuracy = 0.86; F1 score = 0.85; AUC = 0.85) and generalizability (generalization error = 2.95%) in classifying the gait impairment of pwPD compared with speed-matched healthy subjects, using a selected dataset of gait features based on lower trunk acceleration data. Although significantly lower than SVM, tree-based algorithms revealed good classification performances with low generalization errors (RF: test accuracy= 0.86; F1 score = 0.85; AUC = 0.85), and lower computational demand than SVM. ANN was similar to DT in terms of classification metrics but showed significantly higher generalization error (7.26%) than tree-based algorithms and SVM and higher computational demand than the other ML algorithms. Even though KNN showed the fastest time performance, its classification metrics were the lowest. We proposed a feature selection procedure based on the combination of filter, wrapper, embedded, and domain-specific methods that was effective in lowering the risk of overrepresenting multicollinear gait features in the model, resulting in a lower risk of overfitting in the test performances by increasing the explainability of the results at the same time. Because of their accurate results, their simplicity of understanding, and explanability, DT and RF algorithms could represent useful tools for the comprehension of gait disorders by making clinicians participate in the decision process. This is the first time that the accuracy and generalizability of the most performed ML algorithms in classifying pwPD gait abnormalities based on gait data from a single lumbar-mounted IMU have been compared. The findings of this study could be used to incorporate machine learning algorithms into software that processes gait data from lumbar-mounted IMUs. Future research could focus on finding the best tree- based model for classification and prediction problems in gait analysis.

Effective hybrid feature selection using different bootstrap enhances cancers classification performance

BackgroundMachine learning can be used to predict the different onset of human cancers. Highly dimensional data have enormous, complicated problems. One of these is an excessive number of genes plus over-fitting, fitting time, and classification accuracy. Recursive Feature Elimination (RFE) is a wrapper method for selecting the best subset of features that cause the best accuracy. Despite the high performance of RFE, time computation and over-fitting are two disadvantages of this algorithm. Random forest for selection (RFS) proves its effectiveness in selecting the effective features and improving the over-fitting problem.MethodThis paper proposed a method, namely, positions first bootstrap step (PFBS) random forest selection recursive feature elimination (RFS-RFE) and its abbreviation is PFBS- RFS-RFE to enhance cancer classification performance. It used a bootstrap with many positions included in the outer first bootstrap step (OFBS), inner first bootstrap step (IFBS), and outer/ inner first bootstrap step (O/IFBS). In the first position, OFBS is applied as a resampling method (bootstrap) with replacement before selection step. The RFS is applied with bootstrap = false i.e., the whole datasets are used to build each tree. The importance features are hybrid with RFE to select the most relevant subset of features. In the second position, IFBS is applied as a resampling method (bootstrap) with replacement during applied RFS. The importance features are hybrid with RFE. In the third position, O/IFBS is applied as a hybrid of first and second positions. RFE used logistic regression (LR) as an estimator. The proposed methods are incorporated with four classifiers to solve the feature selection problems and modify the performance of RFE, in which five datasets with different size are used to assess the performance of the PFBS-RFS-RFE.ResultsThe results showed that the O/IFBS-RFS-RFE achieved the best performance compared with previous work and enhanced the accuracy, variance and ROC area for RNA gene and dermatology erythemato-squamous diseases datasets to become 99.994%, 0.0000004, 1.000 and 100.000%, 0.0 and 1.000, respectively.ConclusionHigh dimensional datasets and RFE algorithm face many troubles in cancers classification performance. PFBS-RFS-RFE is proposed to fix these troubles with different positions. The importance features which extracted from RFS are used with RFE to obtain the effective features.

Tailoring the density of carbon nanotube networks through chemical self-assembly by click reaction for reliable transistors

Semiconducting single-walled carbon nanotubes (sc-SWNTs) are attracting considerable interest for use as active layers in various electronic applications such as field-effect transistors (FETs) because of their extremely high intrinsic charge carrier mobility and solution processability at low costs. However, it is challenging to achieve a constant sc-SWNT density for ensuring commercial-level, uniform performance in FETs based on random-network SWNT films formed by solution processing. This paper reports a facile method for sorting sc-SWNT and precisely controlling the density of random-network sc-SWNT films by azide-functionalized polymer. The chemical self-assembly of SWNTs is performed between azide-functionalized polymer-wrapped sc-SWNTs and alkyne-based substrate via click reaction. A high-purity sc-SWNT ink is obtained by the conjugated polymer wrapping method using an azide-functionalized polyfluorene in methylcyclohexane. The sc-SWNTs are then chemically bound to a substrate with an alkyne adhesive layer through a Cu-catalyzed azide-alkyne cycloaddition reaction. FETs with dense and uniform SWNT films with a linear density of 30 (±2) tubes μm−1 exhibit markedly high hole mobility of up to 25.4 cm2 V−1 s−1 and excellent performance uniformity. Furthermore, the SWNT films anchored on the substrates are highly resistant to exogenous disruptions, such as sonication in organic solvents, leading the great potential for applications such as biosensors that require strong adhesive strength.

Prediction and optimization of heating and cooling loads for low energy buildings in Morocco: An application of hybrid machine learning methods

In this paper, a metamodel-based approach involving simulation data collection and data-driven techniques was used to forecast and optimize heating and cooling loads in three different climates in Morocco. The metamodel method is gaining popularity as it offers a better balance between accuracy and calculation time. In addition, a wrapper method was used as a feature selection approach to find the best feature subsets in order to reduce models’ complexity. Therefore, the performances of eleven state-of-the-art algorithms including evolutionary-based algorithms, swarm intelligence-based algorithms and human-based algorithms, coupled with a learner such as Artificial Neural Networks (ANNs) and Support vector machine (SVM), were assessed. Hybrid model based on League championship algorithm (LCA), Discrete symbiotic organisms search (DSOS) algorithm, Particle swarm optimization (PSO) algorithm showed better results in terms of accuracy and reduction of feature input parameters. Indeed, the best desired performances were obtained with LCA-SVM for cooling load, DSOS-SVM for heating load, PSO-ANN for both heating and cooling loads. On the other hand, to optimize the annual thermal load, the NSGA-II algorithm was used. Results showed a reduction of 68% of the total annual load compared to the base case in Meknes city, 73% in Ifrane and 67% in Marrakech. The solution that reflects a compromise between the two objective functions (i.e., cooling load and heating load) gives better results in terms of CO2 emissions reduction in all climates evaluated, except in the cold climate.

Classification of Nuclear Reactor Operations Using Spatial Importance and Multisensor Networks

Distributed multisensor networks record multiple data streams that can be used as inputs to machine learning models designed to classify operations relevant to proliferation at nuclear reactors. The goal of this work is to demonstrate methods to assess the importance of each node (a single multisensor) and region (a group of proximate multisensors) to machine learning model performance in a reactor monitoring scenario. This, in turn, provides insight into model behavior, a critical requirement of data-driven applications in nuclear security. Using data collected at the High Flux Isotope Reactor at Oak Ridge National Laboratory via a network of Merlyn multisensors, two different models were trained to classify the reactor’s operational state: a hidden Markov model (HMM), which is simpler and more transparent, and a feed-forward neural network, which is less inherently interpretable. Traditional wrapper methods for feature importance were extended to identify nodes and regions in the multisensor network with strong positive and negative impacts on the classification problem. These spatial-importance algorithms were evaluated on the two different classifiers. The classification accuracy was then improved relative to baseline models via feature selection from 0.583 to 0.839 and from 0.811 ± 0.005 to 0.884 ± 0.004 for the HMM and feed-forward neural network, respectively. While some differences in node and region importance were observed when using different classifiers and wrapper methods, the nodes near the facility’s cooling tower were consistently identified as important—a conclusion further supported by studies on feature importance in decision trees. Node and region importance methods are model-agnostic, inform feature selection for improved model performance, and can provide insight into opaque classification models in the nuclear security domain.

A graph based preordonnances theoretic supervised feature selection in high dimensional data

Generally, for high-dimensional datasets, only some features are relevant, while others are irrelevant or redundant. In the machine learning field, the use of a strategy for eliminating insignificant features from a dataset is very important for the classification task. Feature selection is the process of identifying the most informative features that help in predicting sample classes efficiently in order to achieve better classification performance. In this research paper, a new hybrid feature selection strategy for high-dimensional datasets is proposed to find the most discriminative subset of features for the dataset with the irrelevant and redundant features discarded. The proposed algorithm is called Maximal Clique based on the coefficients Ψ (MaCΨ algorithm). The MaCΨ method has the capability to handle categorical, numerical, and hybrid datasets. Furthermore, it can be applied either to binary or multi-class classification problems. The global structure of the MaCΨ algorithm can be described by three steps. In the first step, a weight is proposed to evaluate the importance of each feature in the dataset by balancing the trade-off between two novel measures of relevance and redundancy, and then the K most important features are selected to form the candidate subset, where K is taken as user input. In the second phase, a wrapper method based on graph theory is applied to the subset retained from the first step to extract the optimal subset of features. In the last stage, the final subset of features with the highest classification performance and the lowest number of features is obtained by applying the backward elimination algorithm to the optimal subset. The performance of the MaCΨ methodology is investigated on artificial as well as real-world datasets with different dimensionalities. The statistical analysis of the experimental results clearly indicates that the MaCΨ approach achieves competitive results in terms of the classification accuracy and the number of selected features compared with some state-of-the-art approaches.

Hybrid Feature Selection of Breast Cancer Gene Expression Microarray Data Based on Metaheuristic Methods: A Comprehensive Review

Breast cancer (BC) remains the most dominant cancer among women worldwide. Numerous BC gene expression microarray-based studies have been employed in cancer classification and prognosis. The availability of gene expression microarray data together with advanced classification methods has enabled accurate and precise classification. Nevertheless, the microarray datasets suffer from a large number of gene expression levels, limited sample size, and irrelevant features. Additionally, datasets are often asymmetrical, where the number of samples from different classes is not balanced. These limitations make it difficult to determine the actual features that contribute to the existence of cancer classification in the gene expression profiles. Various accurate feature selection methods exist, and they are being widely applied. The objective of feature selection is to search for a relevant, discriminant feature subset from the basic feature space. In this review, we aim to compile and review the latest hybrid feature selection methods based on bio-inspired metaheuristic methods and wrapper methods for the classification of BC and other types of cancer.

An Experimental Comprehensive Analysis of Drilling Carbon Fiber Reinforced Plastic Tubes and Comparison With Carbon Fiber Reinforced Plastic Plate

Today, carbon fiber reinforced plastic (CFRP) composites, which are the main areas of use such as aerospace and automotive, can be produced by different methods. Especially in the production of tubular composites, filament winding and roll wrapping methods are used generally, while vacuum-assisted resin transfer molding method or vacuum bagging method comes to the fore in the production of composite plates. These materials produced by different methods may exhibit different behavior in terms of machinability. In this study, the drilling machinability characteristics of CFRP tubes produced by filament winding and roll wrapping methods were investigated and compared with the CFRP plate produced with vacuum bagging and the results were presented comparatively. Experimental results showed that the composite plate generates more force and damage when drilled compared to the tubes, so its machinability is more difficult. Higher thrust force and damage occur in filament wound tube compared to roll wrapped tube. In addition, although the thrust force does not increase with the use of support in composite tubes, a significant improvement in drilling-induced delamination damage behavior is obtained. Compared to the tubes, a better borehole surface quality is obtained in the plate, while more dust chip formation is observed.

Improving Air Pollution Prediction Modelling Using Wrapper Feature Selection

Feature selection is considered as one of the essential steps in data pre-processing. However, all of the previous studies on predicting PM10 concentration in Malaysia have been limited to statistical method feature selection, and none of these studies used machine-learning approaches. Therefore, the objective of this research is to investigate the influence variables of the PM10 prediction model by using wrapper feature selection to compare the prediction model performance of different wrapper feature selection and to predict the concentration of PM10 for the next day. This research uses 10 years of daily data on pollutant concentrations from two stations (Klang and Shah Alam) obtained from the Department of Environment Malaysia (DOE) from 2009 until 2018. Six wrapper methods (forward selection, backward elimination, stepwise, brute-force, weight-guided and genetic algorithm evolution and the predictive analytics multiple linear regression (MLR) and artificial neural network (ANN)) were implemented in this study. This study found that brute-force is the dominant wrapper method in most of the best models in selecting important features for MLR. Moreover, compared to MLR, ANN provides more advantages regarding model accuracy and permits feature selection in predicting PM10. The overall results revealed that the RMSE value for next day prediction in Klang is 20.728, while the AE value is 15.69. Furthermore, the RMSE value for next day prediction in Shah Alam is 10.004, while the AE value is 7.982. Finally, all of the predicted models in Klang and Shah Alam can be used to predict the PM10 concentrations. This proposed model can be used as a tool for an early warning system in giving air quality information to local authorities in order to formulate air-quality-improvement strategies.

Methods and Algorithms for Solving the Problem of Early Diagnostics of Technical Objects Using Data Mining Methods

Русский Main page New issue Archive of articles Editorial board For the authors Publishing house ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES". No. 9. Vol. 28. 2022 DOI: 10.17587/it.28.475-484 G. S. Veresnikov, Dr. Tech. Sc., Leading Researcher, A. V. Golev, Junior Researcher, A. M. Moskovtsev, Engineer, M. P. Martirosyan, Engineer, Institute of Control Sciences V. A. Trapeznikov Academy of Sciences Senior Researcher, Moscow, 117997, Russian Federation Methods and Algorithms for Solving the Problem of Early Diagnostics of Technical Objects Using Data Mining Methods We are seeing a huge increase in the productivity of technical facilities. But as soon as the object is used for its intended purpose, its technical resource begins to be consumed. Early diagnostics and condition detection are important for reliable and safe operation of technical facilities. The results of early diagnostics are used both for prevention of emergency situations and for operation of technical objects according to their condition. In this article, the task is to analyze the methods and algorithms of data mining used in the tasks of determining and predicting the state of technical objects. Neural networks, decision trees, growing pyramid networks, metric algorithms, and the support vector method are considered. Advantages and disadvantages of data mining methods and algorithms are given. Special attention is paid to algorithms and methods for identifying informative features in empirical data, in particular, filtering, projection, and wrapping methods. The use of these methods and algorithms is important to simplify the models used to classify the States of technical objects and improve the accuracy of diagnostics. Analysis of the world scientific literature has shown the relevance of using methods and algorithms for data mining in integration with methods for identifying informative features. These methods are usually used in conjunction with wrapper methods and filters to improve diagnostic accuracy and condition detection of technical objects.

Addressing the Effectiveness of DDoS-Attack Detection Methods Based on the Clustering Method Using an Ensemble Method

The curse of dimensionality, due to lots of network-traffic attributes, has a negative impact on machine learning algorithms in detecting distributed denial of service (DDoS) attacks. This study investigated whether adding the filter and wrapper methods, preceded by combined clustering algorithms using the Vote classifier method, was effective in lowering the false-positive rates of DDoS-attack detection methods. We examined this process to address the curse of dimensionality of machine learning algorithms in detecting DDoS attacks. The results of this study, using ANOVA statistical analyses, showed that incorporating the wrapper method had superior performance in comparison with the filter and clustering methods. IT professionals aim at incorporating effective DDoS-attack detection methods to detect attacks. Therefore, the contribution of this study is that incorporating the wrapper method is the most suitable option for organizations to detect attacks as illustrated in this study. Subsequently, IT professionals could incorporate the DDoS-attack detection methods that, in this study, produced the lowest false-positive rate (0.012) in comparison with all the other mentioned studies.

Using PRISMA Hyperspectral Satellite Imagery and GIS Approaches for Soil Fertility Mapping (FertiMap) in Northern Morocco

Quickly and correctly mapping soil nutrients significantly impact accurate fertilization, food security, soil productivity, and sustainable agricultural development. We evaluated the potential of the new PRISMA hyperspectral sensor for mapping soil organic matter (SOM), available soil phosphorus (P2O5), and potassium (K2O) content over a cultivated area in Khouribga, northern Morocco. These soil nutrients were estimated using (i) the random forest (RF) algorithm based on feature selection methods, including feature subset evaluation and feature ranking methods belonging to three categories (i.e., filter, wrapper, and embedded techniques), and (ii) 107 soil samples taken from the study area. The results show that the RF-embedded method produced better predictive accuracy compared with the filter and wrapper methods. The model for SOM showed moderate accuracy (Rval2 = 0.5, RMSEP = 0.43%, and RPIQ = 2.02), whereas that for soil P2O5 and K2O exhibited low efficiency (Rval2 = 0.26 and 0.36, RMSEP = 51.07 and 182.31 ppm, RPIQ = 0.65 and 1.16, respectively). The interpolation of RF-residuals by ordinary kriging (OK) methods reached the highest predictive results for SOM (Rval2 = 0.69, RMSEP = 0.34%, and RPIQ = 2.56), soil P2O5 (Rval2 = 0.44, RMSEP = 44.10 ppm, and RPIQ = 0.75), and soil K2O (Rval2 = 0.51, RMSEP = 159.29 ppm, and RPIQ = 1.34), representing the best fitting ability between the hyperspectral data and soil nutrients. The result maps provide a spatially continuous surface mapping of the soil landscape, conforming to the pedological substratum. Finally, the hyperspectral remote sensing imagery can provide a new way for modeling and mapping soil fertility, as well as the ability to diagnose nutrient deficiencies.

Investigating the performance of machine learning models combined with different feature selection methods to estimate the energy consumption of buildings

Machine learning is considered a promising method for developing building energy-benchmarking models. However, the high dimensionality of building energy datasets can reduce model accuracy and generalization but increase the computational cost. Meanwhile, the poor interpretability of machine learning models limits the understanding of insights and, in turn, hinders policymaking. Therefore, the first objective of this study was to investigate the benefits of feature selection on the performance of machine learning-based energy usage models. Three typical feature selection methods (filter, wrapper, and embedded) were selected, and the effect of each method was evaluated based on three tree-ensemble learning algorithms. Another objective was to analyze the interpretability of the machine learning model using the Shapley additive explanation method. The results were obtained using a city-scale energy consumption dataset consisting of 478 healthcare buildings in Chongqing, China. It was found that the wrapper method generally improved the accuracy of the machine learning models compared to that of the other two methods. In addition, the model developed using extreme gradient boosting combined with the wrapper method achieved the best accuracy. Moreover, the model interpretability analysis demonstrated important features and revealed how these features influence energy use for individual buildings.

Using end buckles to improve debonding resistance in FRP-bonded precracked RC beams

The strengthening effect of reinforced concrete (RC) members externally bonded (EB) with fiber reinforced polymer (FRP) laminates is significantly affected by interfacial debonding failure. To improve the connection between FRP sheets and concrete, a novel type of anchor device for FRP sheets called buckle was proposed in this study. This device can effectively lock the end of the sheet using a patented method of wrapping. In addition to conventional external bonding, bolts were installed on the buckles to form hybrid anchored (HA) FRP sheet. To verify the effectiveness of this new method, a precracked RC beam was prepared in comparison with two undamaged beams strengthened with EB or HA FRP sheets. As observed, end buckles can limit the development of cracks regardless of initial cracking. Despite the occurrence of intermediate debonding due to initial cracks, both the ultimate capacity and failure ductility were significantly improved. Owing to the end buckles, the debonding resistance was excellent, and the initial cracking of the beam hardly affected the strengthening effect. In addition, existing formulas were used to calculate the lower limit of the ultimate load of the specimen strengthened with HA FRP, which was still higher than that of the specimen strengthened by EB FRP. Therefore, the results confirmed that the debonding resistance was greatly enhanced due to the locking of the FRP sheet via end buckles, and the proposed method is useful to guarantee the strengthening effect.

Lithium Battery Health Factor Extraction Based on Improved Douglas–Peucker Algorithm and SOH Prediction Based on XGboost

To mine the battery’s health factors more comprehensively and accurately identify the lithium battery’s State of Health (SOH), an Improved Douglas–Peucker feature extraction algorithm is proposed, and the LAOS-XGboost model is proposed to be used to predict the SOH of the battery. Firstly, to solve the problem that the traditional Douglas–Peucker algorithm has difficulties extracting curve features in a fixed dimension, the Douglas–Peucker algorithm is improved by de-thresholding. Then, the Wrapper method combined with the Improved Douglas–Peucker algorithm is used to construct the feature engineering of battery life prediction, and the optimal feature subset is obtained. Then, LAOS-XGboost is used to establish a battery SOH prediction model; finally, this model is used to predict the SOH of different batteries and the same battery, and the robustness of the model is analyzed. The experimental results show that the R2 of all XGboost models is higher than 0.97 in the prediction experiments of different batteries. The AE of the LAOS-XGboost model is 0, and the TIC index is less than 3% under 10 dB SNR. In the same battery prediction experiment, the TIC index of the model is less than 0.3%.

A novel hybrid approach for feature selection enhancement: COVID-19 case study

Feature selection is a promising Artificial Intelligence technique for screening, analysing, predicting, and tracking current COVID-19 patients and likely future patients. Significant applications are developed to track data of confirmed, recovered, and death cases. In this work, we propose a new feature selection method based on a new way of hybridization between filter and wrapper methods. The proposed approach is expected to achieve high classification accuracy with a small feature subset. Specifically, the main contribution of this work is a four steps-based approach organized as follows: First, we remove consecutively duplicate and constant features. Then, we select the highest-ranked feature with Mutual Information. In the last step, we run the ‘Backward Feature Elimination’ algorithm to delete features from the active subset until a stopping criterion based on the degradation of classification performance is met. We applied the proposed approach to a COVID-19 dataset to test its ability to find the relevant feature for characterizing the disease, such as new cases infected with the virus, people vaccinated, and the number of deaths, to better assess the situation. For evaluation purposes, experiments are conducted at the first stage on the COVID-19 dataset, then on six benchmark datasets that have a high dimensional and large size. The method performance is tracked and measured on these datasets and a comparison with many approaches is provided.

Classification of diabetic retinopathy with feature selection over deep features using nature-inspired wrapper methods

Diabetic retinopathy (DR) is the most common cause of blindness in middle-aged people. It shows that an automatic image evaluation system is needed in the diagnosis of this disease due to the low number of scans. It is critical to meet this need that these systems are large-scale, cost-effective, and minimally invasive screening programs. With the use of deep learning techniques, it has become possible to develop these systems faster. In this study, a new approach based on feature selection with wrapper methods used for fundus images is presented that can be used for the classification of diabetic retinopathy. The fundus images used in the approach were improved with image processing techniques, thus eliminating unnecessary dark areas in the image. In this new approach, the most effective features are selected by wrapping methods over 512 deep features obtained from EfficientNet and DenseNet models. Binary Bat Algorithm (BBA), Equilibrium Optimizer (EO), Gravity Search Algorithm (GSA), and Gray Wolf Optimizer (GWO) were chosen as wrappers for the proposed approach. Selected features are classified by support vector machines and random forest machine learning methods. Considering the performance of this new approach, it gives the highest value of 96.32 accuracy and 0.98 kappa. These performance values were obtained with a minimum of 250 selected features. The Asia Pacific Tele-Ophthalmology Society (APTOS) dataset used to obtain these values was taken from a competition organized by Kaggle. The highest kappa value in this competition was reported as 0.93. This parameter clearly demonstrates the success of our approach.

Prognosis of Cervical Cancer Disease by Applying Machine Learning Techniques

Cervical cancer is one of the deadliest diseases in women worldwide. It is caused by long-term infection of the skin cells and mucosal cells of the genital area of women. The most disturbing thing about this cancer is the fact that it does not show any symptoms when it occurs. In the diagnosis and prognosis of cervical cancer disease, machine learning has the potential to help detect it at an early stage. In this paper, we analyzed different supervised machine learning techniques to detect cervical cancer at an early stage. To train the machine learning model, a cervical cancer dataset from the UCI repository was used. The different methods were evaluated using this dataset of 858 cervical cancer patients with 36 risk factors and one outcome variable. Six classification algorithms were applied in this study, including an artificial neural network, a Bayesian network, an SVM, a random tree, a logistic tree, and an XG-boost tree. All models were trained with and without a feature selection algorithm to compare the performance and accuracy of the classifiers. Three feature selection algorithms were used, namely (i) relief rank, (ii) wrapper method and (iii) LASSO regression. The maximum accuracy of 94.94% was recorded using XG Boost with complete features. It is also observed that for this dataset, in some cases, the feature selection algorithm performs better. Machine learning has been shown to have advantages over traditional statistical models when it comes to dealing with the complexity of large-scale data and uncovering prognostic features. It offers much potential for clinical use and for improving the treatment of cervical cancer. However, the limitations of prediction studies and models, such as simplified, incomplete information, overfitting, and lack of interpretability, suggest that further efforts are needed to improve the accuracy, reliability, and practicality of clinical outcome prediction.

Correlation-based feature selection using bio-inspired algorithms and optimized KELM classifier for glaucoma diagnosis

Reduced computational time and cost, reduced skilled professional resources, and diagnostic accuracy have made medical diagnosis using computer aided systems (CAD) increasingly popular and can be comfortably employed in the diagnosis of many acute and chronic diseases in ophthalmology, cardiology, cancer detection, etc. There seems to be a growing necessity for the computational algorithms to be robust enough, to identify the abnormality in each of the cases, aiding in early diagnosis. In this paper, a glaucoma diagnostic approach based on the wrapper method employing bio-inspired algorithms, and a Kernel-Extreme Learning Machine (KELM) classifier is proposed. The bio-inspired algorithms are deployed to select feature sub-sets, generating three feature sub-sets from the pre-processed fundus images by adopting a correlation-based feature selection (CFS) approach. The selected features are utilized to train the salp-swarm optimization based KELM, which finds the optimal parameters of the KELM classifier network. The proposed methodology is evaluated on the public and private retinal fundus datasets containing 7280 images. The experimental outcome revealed that the system is able to attain a maximum overall accuracy of 99.61% with 99.89% sensitivity and 100% specificity. A 5-fold cross validation showed 98.78% accuracy ensuring a bias-free classification. Further, by experimenting on degraded images (Gaussian, salt-pepper noise images) of the original dataset, the model achieved extreme robustness with 99.3% accuracy. The proposed method is compared with other similar methods, which showed the efficiency of our method. The framework proposed can aid in making clinical decisions for various pathologies like lung infection, diabetic retinopathy, etc.