Background: The bootstrap can be alternative to cross-validation as a training/test set splitting method since it minimizes the computing time in classification problems in comparison to the tenfold cross-validation. Objectives: Тhis research investigates what proportion should be used to split the dataset into the training and the testing set so that the bootstrap might be competitive in terms of accuracy to other resampling methods. Methods/Approach: Different train/test split proportions are used with the following resampling methods: the bootstrap, the leave-one-out cross-validation, the tenfold cross-validation, and the random repeated train/test split to test their performance on several classification methods. The classification methods used include the logistic regression, the decision tree, and the k-nearest neighbours. Results: The findings suggest that using a different structure of the test set (e.g. 30/70, 20/80) can further optimize the performance of the bootstrap when applied to the logistic regression and the decision tree. For the k-nearest neighbour, the tenfold cross-validation with a 70/30 train/test splitting ratio is recommended. Conclusions: Depending on the characteristics and the preliminary transformations of the variables, the bootstrap can improve the accuracy of the classification problem.

TENFOLD BOOTSTRAP PROCEDURE FOR SUPPORT VECTOR MACHINES

Diabetes causes an increase in the level of blood sugar, which leads to damage to various parts of the human body. Diabetes data are used not only for providing a deeper understanding of the treatment mechanisms but also for predicting the probability that one might become sick. This paper proposes a novel methodology to perform classification in the case of heavy class imbalance, as observed in the PIMA diabetes dataset. The proposed methodology uses two novel steps, namely resampling and random shuffling prior to defining the classification model. The methodology is tested with two versions of cross validation that are appropriate in cases of class imbalance—k-fold cross validation and stratified k-fold cross validation. Our findings suggest that when having imbalanced data, shuffling the data randomly prior to a train/test split can help improve estimation metrics. Our methodology can outperform existing machine learning algorithms and complex deep learning models. Applying our proposed methodology is a simple and fast way to predict labels with class imbalance. It does not require additional techniques to balance classes. It does not involve preselecting important variables, which saves time and makes the model easy for analysis. This makes it an effective methodology for initial and further modeling of data with class imbalance. Moreover, our methodologies show how to increase the effectiveness of the machine learning models based on the standard approaches and make them more reliable.

Purpose: The aim of this research is to propose a modification of the ANOVA-SVM method that can increase accuracy when detecting benign and malignant breast cancer. Methodology: We proposed a new method ANOVA-BOOTSTRAP-SVM. It involves applying the analysis of variance (ANOVA) to support vector machines (SVM) but we use the bootstrap instead of cross validation as a train/test splitting procedure. We have tuned the kernel and the C parameter and tested our algorithm on a set of breast cancer datasets. Findings: By using the new method proposed, we succeeded in improving accuracy ranging from 4.5 percentage points to 8 percentage points depending on the dataset. Research limitations: The algorithm is sensitive to the type of kernel and value of the optimization parameter C. Practical implications: We believe that the ANOVA-BOOTSTRAP-SVM can be used not only to recognize the type of breast cancer but also for broader research in all types of cancer. Originality/value: Our findings are important as the algorithm can detect various types of cancer with higher accuracy compared to standard versions of the Support Vector Machines.

Feature selection is a powerful tool to identify the important characteristics of data for prediction. Feature selection, therefore, can be a tool for avoiding overfitting, improving prediction accuracy and reducing execution time. The applications of feature selection procedures are particularly important in Support vector machines, which is used for prediction in large datasets. The larger the dataset, the more computationally exhaustive and challenging it is to build a predictive model using the support vector classifier. This paper investigates how the feature selection approach based on the analysis of variance (ANOVA) can be optimized for Support Vector Machines (SVMs) to improve its execution time and accuracy. We introduce new conditions on the SVMs prior to running the ANOVA to optimize the performance of the support vector classifier. We also establish the bootstrap procedure as alternative to cross validation to perform model selection. We run our experiments using popular datasets and compare our results to existing modifications of SVMs with feature selection procedure. We propose a number of ANOVA-SVM modifications which are simple to perform, while at the same time, boost significantly the accuracy and computing time of the SVMs in comparison to existing methods like the Mixed Integer Linear Feature Selection approach.

This study's objective is to offer a practical computer method for handling classification problems on large datasets. The aim of this study is to offer a practical computer approach for handling classification tasks on big datasets. We show that using Python’s built-in parameters to balance classes can improve the accuracy and the metrics of a classification task. We employ logistic regression, support vector machines, decision trees, and random forest classifier. We use the parameter “class_weight='balanced'” to run each classification model as well as stratified train/test splitting to ensure that relative class frequencies are approximately preserved in each train and set subsets. We use our methods on medical datasets because class imbalance is frequently a problem there. Our research shows that the proposed algorithms can improve the accuracy and classification metrics of the given medical datasets. We propose an effective and easy-to-apply alternative to improve the prediction ability of the presented classification models in medical datasets. We test an easily reproducible set-up where any classification model can be used to model imbalanced classes. The key tuning of the model lies in the stratified train/test split as well as the parameter “class weight='balanced'”. By combination of parameter tuning, better classification performance can be obtained in a quick and simple manner. It is simple and quick to replicate our algorithms to examine various medical datasets and determine which model best fits the data. It can be reproduced in biostatistical laboratories and by medical companies. Because it is simple to comprehend, medical researchers can swiftly review the information and determine the best course of action.