Performance Of Machine Learning Classifiers Research Articles

Applications of Text Mining techniques to extract meaningful information from gastroenterology medical reports

Text mining techniques, particularly topic modeling, can be used for the automatic extraction of information from medical reports. The ability to autonomously analyze texts and identify topics within them can provide meaningful clinical insights that support physicians in diagnostic settings and enhance the characterization of intestinal diseases, leading to more efficient and automated systems.This study evaluates the effectiveness of Latent Dirichlet Allocation (LDA) and BERTopic in modeling topics from colonoscopy reports related to Crohn’s Disease, Ulcerative Colitis, and Polyps. We compared these models in terms of their ability to identify clinically relevant topics, their influence on the performance of machine learning classifiers trained on the derived topic features, and their scalability.Our analysis, based on average results across five iterations of train-test splits, showed that BERTopic generally outperformed LDA in clustering metrics, achieving Adjusted Rand Index (ARI), Normalized Mutual Information (NMI), and Purity scores of 0.5637, 0.5953, and 0.8447, respectively, compared to LDA’s scores of 0.5349, 0.5254, and 0.8149. Additionally, classifiers trained on BERTopic-derived features exhibited improved predictive accuracy and F1-scores, with Logistic Regression reaching a mean accuracy of 0.8464 and a mean F1-score of 0.8507, compared to 0.8319 and 0.8351 for LDA-based features. Despite BERTopic’s overall superior performance, LDA demonstrated greater stability and interpretability, making it a viable option in scenarios where computational efficiency is a priority.

The effect of data complexity on classifier performance

The research area of Software Defect Prediction (SDP) is both extensive and popular, and is often treated as a classification problem. Improvements in classification, pre-processing and tuning techniques, (together with many factors which can influence model performance) have encouraged this trend. However, no matter the effort in these areas, it seems that there is a ceiling in the performance of the classification models used in SDP. In this paper, the issue of classifier performance is analysed from the perspective of data complexity. Specifically, data complexity metrics are calculated using the Unified Bug Dataset, a collection of well-known SDP datasets, and then checked for correlation with the defect prediction performance of machine learning classifiers (in particular, the classifiers C5.0, Naive Bayes, Artificial Neural Networks, Random Forests, and Support Vector Machines). In this work, different domains of competence and incompetence are identified for the classifiers. Similarities and differences between the classifiers and the performance metrics are found and the Unified Bug Dataset is analysed from the perspective of data complexity. We found that certain classifiers work best in certain situations and that all data complexity metrics can be problematic, although certain classifiers did excel in some situations.

Improving and comparing performance of machine learning classifiers optimized by swarm intelligent algorithms for code smell detection

In complex systems, the maintenance phase engenders the emergence of code smells due to incessant shifts in requirements and designs, stringent timelines, and the developer's relative inexperience. While not conventionally classified as errors, code smells inherently signify flawed design structures that lead to future bugs and errors. It increases the software budget and eventually makes the system hard to maintain or completely obsolete. To mitigate these challenges, practitioners must detect and refactor code smells. However, the theoretical interpretation of smell definitions and intelligent establishment of threshold values pose a significant conundrum. Supervised machine learning emerges as a potent strategy to address these problems and alleviate the dependence on expert intervention. The learning mechanism of these algorithms can be refined through data pre-processing and hyperparameter tuning. Selecting the best values for hyperparameters can be tedious and requires an expert. This study introduces an innovative paradigm that fuses twelve swarm-based, meta-heuristic algorithms with two machine learning classifiers, optimizing their hyperparameters, eliminating the need for an expert, and automating the entire code smell detection process. Through this synergistic approach, the highest post-optimization accuracy, precision, recall, F-measure, and ROC-AUC values are 99.09%, 99.20%, 99.09%, 98.06%, and 100%, respectively. The most remarkable upsurge is 35.9% in accuracy, 53.79% in precision, 35.90% in recall, 44.73% in F-measure, and 36.28% in ROC-AUC. Artificial Bee Colony, Grey Wolf, and Salp Swarm Optimizer are the top-performing swarm-intelligent algorithms. God and Data Class are the most readily detectable smells with optimized classifiers. Statistical tests underscore the profound impact of employing swarm-based algorithms to optimize machine learning classifiers, corroborated by statistical tests. This seamless integration enhances classifier performance, automates code smell detection, and offers a robust solution to a persistent software engineering challenge.

Enhanced Brain Tumor Classification: A Hybrid Classifier Approach

Early detection of brain tumors is crucial for enhancing patient survival rates and treatment effectiveness. Analyzing magnetic resonance imaging (MRI) images by manually is a difficult task. Therefore, there is a requirement for digital techniques to improve the accuracy of tumor diagnosis. Study of this work was to enhance the performance of Machine Learning classifiers rather than move to deep learning. Machine Learning classifiers are advantageous compared to deep learning methods because of their ability to train on small datasets, low computing time complexity, affordability, and ease of adoption by individuals with less expertise The proposed methodology encompasses image processing, segmentation, feature extraction, feature reduction, and classification, leveraging techniques such as Otsu’s thresholding, GLCM, BWT, and PCA to extract pertinent information while minimizing dimensionality. By integrating a hybrid classifier that combines SVM, DT, and KNN, along with ensemble techniques, significant improvements in classification accuracy and robustness are achieved. Experimental results on the TCGA-GBM dataset showcase accuracy of 96.9%, with high specificity (95.14%) and sensitivity (97.14%) for classify normal tumors (benign tumors) and cancerous tumors (malignant tumors). Furthermore, the proposed methodology demonstrates efficiency with relatively low execution time, rendering it suitable for real-world applications.

Retracted

Retracted

Exploring Stress Detection in Tweets: A Comparative Business Analysis of Classification Algorithms

Abstract: In today's digitally-driven landscape, social media platforms such as Twitter, Facebook, and WordPress have emerged as critical conduits for public discourse, resulting in an overwhelming influx of unprocessed data, particularly on Twitter. Coping with the sheer volume of this data presents a significant challenge. To address this, sentiment analysis, a technique that categorizes sentiments into positive, negative, or neutral, offers a promising solution. This study delves into three main approaches for sentiment analysis: Machine learning-based methods, Sentiment lexicon-based approaches, and Hybrid methods. Its primary research objectives revolve around the identification of suitable algorithms and metrics for evaluating the performance of Machine Learning Classifiers. Additionally, the study aims to compare these metrics with respect to the dataset size, gauging their impact on the most appropriate sentiment analysis algorithm.The research methodology applied is experimental, entailing a rigorous assessment of algorithms using carefully selected metrics. The results of this investigation spotlight Naïve Bayes, Random Forest, XGBoost, and CNN-LSTM as the leading machine learning algorithms under consideration. These algorithms are assessed based on key performance metrics such as precision, accuracy, F1 score, and recall. Notably, the CNN-LSTM model emerges as the optimal choice for sentiment analysis of Twitter data within the specified dataset size, achieving a remarkable accuracy rate of 88%.In summary, this research successfully pinpoints the most suitable algorithm for sentiment analysis of Twitter data, especially in the context of dataset size. The CNN- LSTM model showcases its effectiveness, serving as a robust tool for sentiment analysis and delivering an impressive accuracy rate. This study significantly enhances our comprehension of public sentiment on Twitter, providing valuable insights into the ever-evolving realm of digital discourse and the analysis of vast unprocessed data

Revolutionizing gastric cancer diagnosis through advanced machine learning approaches

<p class="Keywords">Early detection of gastric cancer through a Computer-Aided Detection (CAD) system has the potential to significantly reduce the mortality rate associated with this disease. This study aims to investigate the effects of class imbalance on the performance of machine learning classifiers in this context. Using a dataset of 145,787 screening records from NHS Liverpool Hospital, we employed stratified sampling to create balanced and unbalanced datasets and evaluated the performance of four machine learning algorithms—Logistic Regression, Support Vector Machine, Naive Bayes, and Multilayer Perceptron—under five different test conditions. The study’s novelty lies in its detailed examination of class imbalance in gastric cancer diagnosis, emphasizing the crucial role of balanced datasets in machine learning-based early detection systems. For the MLP model under 10-fold cross-validation, the Class 0 sensitivity (non-cancer cases) of the unbalanced dataset was 0.968, higher than the balanced dataset’s 0.902. However, the Class 1 sensitivity (cancer cases) and Positive Predictive Value (PPV) of the unbalanced dataset were much lower (0.383 and 0.527) than those of the balanced dataset (0.959 and 0.907), indicating a significant improvement in identifying true positive cases when using a balanced dataset. These findings highlight the negative effect of class imbalance on prediction accuracy for positive cancer cases and underscore the importance of addressing this imbalance for more reliable and accurate predictions in medical diagnosis and screening. This approach has the potential to improve patient outcomes and may contribute to strategies aimed at reducing the mortality rate associated with gastric cancer.</p>

Analysis of Machine Learning Classifiers for Speaker Identification: A Study on SVM, Random Forest, KNN, and Decision Tree

This study investigates the performance of machine learning classifiers in the domain of speaker identification, a pivotal component of modern digital security systems. With the burgeoning integration of voice-activated interfaces in technology, the demand for accurate and reliable speaker identification is paramount. This research provides a comprehensive comparison of four widely used classifiers: Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Decision Tree (DT). Utilizing the LibriSpeech dataset, known for its diversity of speakers and recording conditions, we extracted Mel-frequency cepstral coefficients (MFCCs) to serve as features for training and evaluating the classifiers. Each model's performance was assessed based on precision, recall, F1-score, and accuracy. The results revealed that RF outperformed all other classifiers, achieving near-perfect metrics, indicative of its robustness and generalizability for speaker identification tasks. KNN also demonstrated high performance, suggesting its suitability for applications where rapid execution and interpretability are critical. Conversely, SVM and DT, while yielding moderate and lower performances respectively, highlighted the necessity for further optimization. These findings underscore the effectiveness of ensemble and distance-based classifiers in handling complex patterns for speaker differentiation. The study not only guides the selection of appropriate classifiers for speaker identification but also sets the stage for future research, which could explore hybrid models and the impact of dataset variability on performance. The insights from this analysis contribute significantly to the field, providing a benchmark for developing advanced speaker identification systems

Boosting and Comparing Performance of Machine Learning Classifiers with Meta-heuristic Techniques to Detect Code Smell

Background: Continuous modifications, suboptimal software design practices, and stringent project deadlines contribute to the proliferation of code smells. Detecting and refactoring these code smells are pivotal to maintaining complex and essential software systems. Neglecting them may lead to future software defects, rendering systems challenging to maintain, and eventually obsolete. Supervised machine learning techniques have emerged as valuable tools for classifying code smells without needing expert knowledge or fixed threshold values. Further enhancement of classifier performance can be achieved through effective feature selection techniques and the optimization of hyperparameter values. Aim: Performance measures of multiple machine learning classifiers are improved by fine tuning its hyperparameters using various type of meta-heuristic algorithms including swarm intelligent, physics, math, and bio-based etc. Their performance measures are compared to find the best meta-heuristic algorithm in the context of code smell detection and its impact is evaluated based on statistical tests. Method: This study employs sixteen contemporary and robust meta-heuristic algorithms to optimize the hyperparameters of two machine learning algorithms: Support Vector Machine (SVM) and k-nearest Neighbors (k-NN). The No Free Lunch theorem underscores that the success of an optimization algorithm in one application may not necessarily extend to others. Consequently, a rigorous comparative analysis of these algorithms is undertaken to identify the best-fit solutions for code smell detection. A diverse range of optimization algorithms, encompassing Arithmetic, Jellyfish Search, Flow Direction, Student Psychology Based, Pathfinder, Sine Cosine, Jaya, Crow Search, Dragonfly, Krill Herd, Multi-Verse, Symbiotic Organisms Search, Flower Pollination, Teaching Learning Based, Gravitational Search, and Biogeography-Based Optimization, have been implemented. Results: In the case of optimized SVM, the highest attained accuracy, AUC, and F-measure values are 98.75%, 100%, and 98.57%, respectively. Remarkably, significant increases in accuracy and AUC, reaching 32.22% and 45.11% respectively, are observed. For k-NN, the best accuracy, AUC, and F-measure values are all perfect at 100%, with noteworthy hikes in accuracy and ROC-AUC values, amounting to 43.89% and 40.83%, respectively. Conclusion: Optimized SVM exhibits exceptional performance with the Sine Cosine Optimization algorithm, while k-NN attains its peak performance with the Flower Optimization algorithm. Statistical analysis underscores the substantial impact of employing meta-heuristic algorithms for optimizing machine learning classifiers, enhancing their performance significantly. Optimized SVM excels in detecting the God Class, while optimized k-NN is particularly effective in identifying the Data Class. This innovative fusion automates the tuning process and elevates classifier performance, simultaneously addressing multiple longstanding challenges.

Enhancing landslide susceptibility modelling through a novel non-landslide sampling method and ensemble learning technique

In recent years, several catastrophic landslide events have been observed throughout the globe, threatening to lives and infrastructures. To minimize the impact of landslides, the need of landslide susceptibility map is important. The study aims to extract high-quality non-landslide samples and improve the accuracy of landslide susceptibility modelling (LSM) outcomes by applying a coupled method of ensemble learning and Machine Learning (ML). The Zigui-Badong section of the Three Gorges Reservoir area (TGRA) in China was considered in the present study. Twelve influencing factors were selected as inputs for LSM, and the relationship between each causal factor and landslide spatial development was quantitatively analyzed. A total of 179 landslides have been used in the present study. About 70% of the landslide pixels were randomly considered for training, and the remaining 30% were used for validation. Logistic Regression (LR) model was applied to produce an initial susceptibility map, and the non-landslide samples were selected within the classified low-susceptibility zone. Subsequently, two ML classifiers – the Classification and Regression Tree (CART), and the Multi-Layer Perceptron (MLP), and four coupling models – the CART-Bagging, CART-Boosting, MLP-Bagging, and MLP-Boosting, were utilized for LSM. Finally, the receiver operating characteristics (ROC) curve and statistical analysis were applied for accuracy assessment. The results show that altitude and distance to rivers were the main causal factors of landslides in the study area. The LR-MLP-Boosting performed the best with an accuracy of 0.986 followed by the LR-CART-Bagging, LR-CART-Boosting, and LR-MLP-Bagging. Accuracy comparisons demonstrate that ensemble learning algorithm can notably enhance the LSM performance of ML classifiers, and the Boosting algorithm marginally outperforms the Bagging algorithm. Moreover, the LR model can effectively constrain the selection range of non-landslide samples. The non-landslide sampling method constrained by LR yields higher quality samples compared to raditional random sampling method with no constraints, which develops a more excellent LSM.

Enhancing Parkinson's disease severity assessment through voice-based wavelet scattering, optimized model selection, and weighted majority voting

Parkinson's disease (PD) is a neurodegenerative disorder characterized by motor and non-motor symptoms that significantly impact an individual's quality of life. Voice changes have shown promise as early indicators of PD, making voice analysis a valuable tool for early detection and intervention. This study aims to assess and detect the severity of PD through voice analysis using the mobile device voice recordings dataset. The dataset consisted of recordings from PD patients at different stages of the disease and healthy control subjects. A novel approach was employed, incorporating a voice activity detection algorithm for speech segmentation and the wavelet scattering transform for feature extraction. A Bayesian optimization technique is used to fine-tune the hyperparameters of seven commonly used classifiers and optimize the performance of machine learning classifiers for PD severity detection. AdaBoost and K-nearest neighbor consistently demonstrated superior performance across various evaluation metrics among the classifiers. Furthermore, a weighted majority voting (WMV) technique is implemented, leveraging the predictions of multiple models to achieve a near-perfect accuracy of 98.62%, improving classification accuracy. The results highlight the promising potential of voice analysis in PD diagnosis and monitoring. Integrating advanced signal processing techniques and machine learning models provides reliable and accessible tools for PD assessment, facilitating early intervention and improving patient outcomes. This study contributes to the field by demonstrating the effectiveness of the proposed methodology and the significant role of WMV in enhancing classification accuracy for PD severity detection.

Heart Failure prediction on diversified datasets to improve generalizability using 2-Level Stacking

Heart disease is a leading cause of death worldwide, and early detection is crucial for improving patient outcomes. Machine learning classifiers have shown promise in predicting heart disease using patient-specific factors such as demographic information, medical history, and lifestyle habits. In this paper, we aimed to evaluate the performance of machine learning classifiers in predicting heart disease using a combination of 5 different datasets such as Cleveland, Hungarian, Switzerland, Long Beach VA, and StatLog (Heart) Datasets available on IEEE data port. The two significant challenges are addressed in this work: 1) predicting heart failure using machine-learning models without eliminating any clinical features, which increases the risk of overfitting and can result in poor performance metrics, and 2) we propose a model that will provide remarkable accuracy regardless of the type of data, to offer model generalizability. Machine-learning algorithms such as logistic regression, decision trees, random forests, support vector machine, extreme gradient boosting, extra-tree, and K-nearest neighbor are applied for heart disease prediction. In addition, ensemble approaches such as majority voting, boosting, bagging, and stacking are employed. The performance of the classifiers is evaluated using performance metrics such as accuracy, precision, recall, and F1-score. The results showed that the ensemble approach of stacking outperformed individual models, with an accuracy of 93.67%.

Automated assessment of foot elevation in adults with hereditary spastic paraplegia using inertial measurements and machine learning

BackgroundHereditary spastic paraplegias (HSPs) cause characteristic gait impairment leading to an increased risk of stumbling or even falling. Biomechanically, gait deficits are characterized by reduced ranges of motion in lower body joints, limiting foot clearance and ankle range of motion. To date, there is no standardized approach to continuously and objectively track the degree of dysfunction in foot elevation since established clinical rating scales require an experienced investigator and are considered to be rather subjective. Therefore, digital disease-specific biomarkers for foot elevation are needed.MethodsThis study investigated the performance of machine learning classifiers for the automated detection and classification of reduced foot dorsiflexion and clearance using wearable sensors. Wearable inertial sensors were used to record gait patterns of 50 patients during standardized 4 times 10 m walking tests at the hospital. Three movement disorder specialists independently annotated symptom severity. The majority vote of these annotations and the wearable sensor data were used to train and evaluate machine learning classifiers in a nested cross-validation scheme.ResultsThe results showed that automated detection of reduced range of motion and foot clearance was possible with an accuracy of 87%. This accuracy is in the range of individual annotators, reaching an average accuracy of 88% compared to the ground truth majority vote. For classifying symptom severity, the algorithm reached an accuracy of 74%.ConclusionHere, we show that the present wearable gait analysis system is able to objectively assess foot elevation patterns in HSP. Future studies will aim to improve the granularity for continuous tracking of disease severity and monitoring therapy response of HSP patients in a real-world environment.

Non-contrast CT synthesis using patch-based cycle-consistent generative adversarial network (Cycle-GAN) for radiomics and deep learning in the era of COVID-19

Handcrafted and deep learning (DL) radiomics are popular techniques used to develop computed tomography (CT) imaging-based artificial intelligence models for COVID-19 research. However, contrast heterogeneity from real-world datasets may impair model performance. Contrast-homogenous datasets present a potential solution. We developed a 3D patch-based cycle-consistent generative adversarial network (cycle-GAN) to synthesize non-contrast images from contrast CTs, as a data homogenization tool. We used a multi-centre dataset of 2078 scans from 1,650 patients with COVID-19. Few studies have previously evaluated GAN-generated images with handcrafted radiomics, DL and human assessment tasks. We evaluated the performance of our cycle-GAN with these three approaches. In a modified Turing-test, human experts identified synthetic vs acquired images, with a false positive rate of 67% and Fleiss’ Kappa 0.06, attesting to the photorealism of the synthetic images. However, on testing performance of machine learning classifiers with radiomic features, performance decreased with use of synthetic images. Marked percentage difference was noted in feature values between pre- and post-GAN non-contrast images. With DL classification, deterioration in performance was observed with synthetic images. Our results show that whilst GANs can produce images sufficient to pass human assessment, caution is advised before GAN-synthesized images are used in medical imaging applications.

A Machine Learning Method with Hybrid Feature Selection for Improved Credit Card Fraud Detection

With the rapid developments in electronic commerce and digital payment technologies, credit card transactions have increased significantly. Machine learning (ML) has been vital in analyzing customer data to detect and prevent fraud. However, the presence of redundant and irrelevant features in most real-world credit card data degrades the performance of ML classifiers. This study proposes a hybrid feature-selection technique consisting of filter and wrapper feature-selection steps to ensure that only the most relevant features are used for machine learning. The proposed method uses the information gain (IG) technique to rank the features, and the top-ranked features are fed to a genetic algorithm (GA) wrapper, which uses the extreme learning machine (ELM) as the learning algorithm. Meanwhile, the proposed GA wrapper is optimized for imbalanced classification using the geometric mean (G-mean) as the fitness function instead of the conventional accuracy metric. The proposed approach achieved a sensitivity and specificity of 0.997 and 0.994, respectively, outperforming other baseline techniques and methods in the recent literature.

From modeling dose-response relationships to improved performance of decision-tree classifiers for predictive toxicology of nanomaterials

The development and application of predictive models towards toxicity of engineered nanomaterials is still far from being satisfactory. One promising contribution to confront this challenge is to effectively augment the performance of machine learning classifiers by progressing the approach towards balancing experimental toxicity data. We propose an improved balancing methodology by fitting the in-vitro toxicological dose-response datasets of engineered nanomaterials to three, four, and five, free parameter dose-response models. The four-free parameter model displays the best fit (in terms of adjusted R2) for most of the examined data. The fitted curve yields, in each case, a continuous sequence of data points, which extends the restricted experimental data and generates additional fitted data points for the minority class, leading to the formation of balanced data for predicting the nanoparticle’s toxicology by decision tree classifiers. The ability to best predict the experimental toxicity data, by applying the decision tree model, was tested by forming three versions of the same experimental data: the imbalanced raw experimental data, the balanced data by applying the common Synthetic Minority Oversampling Technique, and by using the approach of Balanced Fitted Dose-Response method, introduced in the present study. We demonstrate that our approach provides improved performance of decision trees in predicting nanoparticles’ toxicity, a method that pertains also to chemical toxicity, central in health and environmental research.

Strain FBG-Based Sensor for Detecting Fence Intruders Using Machine Learning and Adaptive Thresholding.

This paper demonstrates an intruder detection system using a strain-based optical fiber Bragg grating (FBG), machine learning (ML), and adaptive thresholding to classify the intruder as no intruder, intruder, or wind at low levels of signal-to-noise ratio. We demonstrate the intruder detection system using a portion of a real fence manufactured and installed around one of the engineering college's gardens at King Saud University. The experimental results show that adaptive thresholding can help improve the performance of machine learning classifiers, such as linear discriminant analysis (LDA) or logistic regression algorithms in identifying an intruder's existence at low optical signal-to-noise ratio (OSNR) scenarios. The proposed method can achieve an average accuracy of 99.17% when the OSNR level is <0.5 dB.

Evaluating the Performance of Machine Learning Classifiers for Detecting Twitter Spam

Evaluating the Performance of Machine Learning Classifiers for Detecting Twitter Spam

Performance of machine learning algorithms for dementia assessment: impacts of language tasks, recording media, and modalities

ObjectivesAutomatic speech and language assessment methods (SLAMs) can help clinicians assess speech and language impairments associated with dementia in older adults. The basis of any automatic SLAMs is a machine learning (ML) classifier that is trained on participants’ speech and language. However, language tasks, recording media, and modalities impact the performance of ML classifiers. Thus, this research has focused on evaluating the effects of the above-mentioned factors on the performance of ML classifiers that can be used for dementia assessment.MethodologyOur methodology includes the following steps: (1) Collecting speech and language datasets from patients and healthy controls; (2) Using feature engineering methods which include feature extraction methods to extract linguistic and acoustic features and feature selection methods to select most informative features; (3) Training different ML classifiers; and (4) Evaluating the performance of ML classifiers to investigate the impacts of language tasks, recording media, and modalities on dementia assessment.ResultsOur results show that (1) the ML classifiers trained with the picture description language task perform better than the classifiers trained with the story recall language task; (2) the data obtained from phone-based recordings improves the performance of ML classifiers compared to data obtained from web-based recordings; and (3) the ML classifiers trained with acoustic features perform better than the classifiers trained with linguistic features.ConclusionThis research demonstrates that we can improve the performance of automatic SLAMs as dementia assessment methods if we: (1) Use the picture description task to obtain participants’ speech; (2) Collect participants’ voices via phone-based recordings; and (3) Train ML classifiers using only acoustic features. Our proposed methodology will help future researchers to investigate the impacts of different factors on the performance of ML classifiers for assessing dementia.

Deriving quantitative information from multiparametric MRI via Radiomics: Evaluation of the robustness and predictive value of radiomic features in the discrimination of low-grade versus high-grade gliomas with machine learning

PurposeAnalysis pipelines based on the computation of radiomic features on medical images are widely used exploration tools across a large variety of image modalities. This study aims to define a robust processing pipeline based on Radiomics and Machine Learning (ML) to analyze multiparametric Magnetic Resonance Imaging (MRI) data to discriminate between high-grade (HGG) and low-grade (LGG) gliomas. MethodsThe dataset consists of 158 multiparametric MRI of patients with brain tumor publicly available on The Cancer Imaging Archive, preprocessed by the BraTS organization committee. Three different types of image intensity normalization algorithms were applied and 107 features were extracted for each tumor region, setting the intensity values according to different discretization levels. The predictive power of radiomic features in the LGG versus HGG categorization was evaluated by using random forest classifiers. The impact of the normalization techniques and of the different settings in the image discretization was studied in terms of the classification performances. A set of MRI-reliable features was defined selecting the features extracted according to the most appropriate normalization and discretization settings. ResultsThe results show that using MRI-reliable features improves the performance in glioma grade classification (AUC=0.93±0.05) with respect to the use of raw (AUC=0.88±0.08) and robust features (AUC=0.83±0.08), defined as those not depending on image normalization and intensity discretization. ConclusionsThese results confirm that image normalization and intensity discretization strongly impact the performance of ML classifiers based on radiomic features. Thus, special attention should be provided in the image preprocessing step before typical radiomic and ML analysis are carried out.