In streaming data environments, drift is observed as changes in the statistical properties of input features. Detection of such phenomena is important because it affects the performance of machine learning models. Traditional drift detection methods often focus on error rates or output distributions, neglecting the root cause: changes in feature importance. This study proposes a drift detection method based on feature ranking in neighboring windows. Feature ranking can be performed using different methods, which will be discussed in the paper. The performance of the drift detector also depends on the classifier used. This work analyzes drifts of different speeds: sudden, gradual, recurrent, and incremental. Experiments on real-world datasets(balanced and imbalanced) show that the proposed method is effective. The paper also provides guidelines for dynamic data environments.

Hidden changes in the data stream are unknown to learning algorithms and are referred to in the literature as drifts of various types. The accuracy of the classifier may degrade due to the occurrence of drift in non-stationary data streams. In such situations, the classifier must detect significant data changes and adjust its predictions. This article aims to present a new method of drift detection based on analyzing changes in feature ranks across adjacent chunks of data. The proposed strategy involves determining the ranking of the most important feature and tracking its fluctuations within the chunks into which the input data stream is divided. Changes in feature rankings between adjacent chunks serve as symptoms of data drift. The Least Absolute Shrinkage and Selection Operator (LASSO) procedure was proposed as an efficient rank pointer. We compared well-known and popular drift detection algorithms, such as the Drift Detection Method (DDM), Early Drift Detection Method (EDDM), ADaptive WINdowing (ADWIN), and Principal Component Analysis Feature Drift Detection (PCA-FDD), with our approach in comparative studies. The tests were conducted on different artificial data streams (sudden, gradual, recurring, and incremental) as well as real data. Comparative studies were performed on both two-class and multi-class datasets. The experiments confirm that the proposed feature drift detection strategy produces valuable results.

Feature drift is a subtype of data distribution drift that occurs when the statistical significance of input features changes over time, despite the overall decision boundary remaining stable. This phenomenon can cause a subtle degradation in model accuracy in streaming environments. In this paper, we propose a new model-aware method called feature importance-driven drift detection (FIDD). Rather than relying on classification error signals, FIDD tracks changes in feature importance rankings obtained from LASSO regression across neighbouring data fragments. As it observes the dynamics of feature importance instead of global label shifts, this method is particularly suited to detecting subtle shifts in data distribution. Experimental evaluation on both synthetic and real-world data (including different types of drift, such as abrupt, gradual, incremental and recurrent) shows that FIDD achieves higher accuracy consistently and produces significantly fewer false alarms than standard drift detectors (e.g., DDM, EDDM and ADWIN). Furthermore, FIDD is robust to labelling noise and computationally efficient, which makes it a practical and interpretable solution for adaptive learning in real-time applications.

The effectiveness of ranking methods in detecting feature drifts in data streams has been evaluated in this work. The FBDD (Feature-Based Drift Detector) method was used for feature drift detection, as creating a feature ranking is a key component of this method. The study was conducted on artificial and real datasets, representing various drifts, such as abrupt, gradual, incremental, and recurring changes. Ten widely used ranking methods were evaluated, including LASSO, the Laplacian Score, and the Kolmogorov-Smirnov method. The analysis focused on key metrics such as classification accuracy (ACC), Matthews correlation coefficient (MCC), and computational efficiency, providing a comprehensive overview of the strengths and weaknesses of each method. The results revealed significant differences in the performance of the methods depending on the nature of the data and the type of drift. This work provides insights into the practical applications of ranking methods for drift detection. It highlights the trade-offs between accuracy, computational efficiency, and the ability to handle different drifts. The findings aim to support researchers and practitioners in selecting the most suitable methods for specific data streams.

In recent years, a Concept Drift topic have gained popularity in scientific papers. Most papers, however, concentrate on describing new techniques for detecting this phenomenon without giving clear use cases or explaining how to use such detectors in a data stream pipeline. In this paper, we propose a complete toolkit for utilizing various types of Concept Drift detectors and visualizing their outcome. Additionally, we propose some metrics that could be suitable for comparing the effectiveness of various detectors.

The article presents a new method of multibiometric Verification that has been enhanced with dynamic classifier selection based on determining their competence. The competence of a classifier is defined, taking into account the type of biometric trait and the samples analyzed. The proposed approach not only allows for adaptive selection of classifiers to specific features and samples but also increases system efficiency by more effectively utilizing models that best match the given conditions. Classifier selection occurs dynamically, which enables the system to adjust flexibly to changing conditions and leverage the strengths of individual models while mitigating their weaknesses through compensation within the committee. The effectiveness of this method has been verified through experiments, which confirmed a significant improvement in person Verification effectiveness compared to methods without dynamic classifier selection.

Data stream analysis is a rapidly growing field driven by the prevalence of massive, dynamic datasets across various domains, including science, medicine, biometrics, and technology. Many data streams are inherently non-homogeneous, highly variable over time, and non-stationary, presenting unique challenges for machine learning applications. This paper addresses the underexplored phenomenon of feature drift—a type of drift where the importance of certain features diminishes while others become more relevant for predictive modeling. Feature drift necessitates continuous monitoring and adaptation of classification models to ensure high-quality predictions. We propose a novel feature drift detection algorithm that operates flexibly in supervised and unsupervised modes. The algorithm is evaluated using synthetic and real-world datasets, demonstrating its ability to effectively detect and adapt to feature drift. Additionally, we investigate various drift scenarios and assess the algorithm’s performance across multiple features. Comparative analyses with state-of-the-art methods highlight the strengths of our approach in maintaining robust classification accuracy and reducing false detections. This research contributes to advancing adaptive machine-learning techniques for dynamic, real-world environments.

The concept and data drift problems have received much attention in recent years. This aspect is crucial in many domains exhibiting non-stationary and cyclical patterns affecting their generative processes. Drift detection can be treated as a supervised task, with labeled data constantly used to validate the learned model. From a practical point of view, this is an impractical task because labeling is complex, costly, and time-consuming. On the other hand, unsupervised change detection techniques are cumbersome in applications because they generate many false alarms. The paper presents a new concept drift detection method based on feature analysis. Stream of data carries information about the distribution patterns that reflect different concepts that may be hidden in the data. The essential features are searched and ranked by LASSO. The rank of features and statistics are employed to feature drift detection. The proposed approach was experimentally checked based on synthetic and natural datasets. The results show that the proposed FBDD algorithm has an advantage over other solutions.

In this paper, we propose practical Python programs together with appropriate environments for the analysis of data stream drift, including the analysis of feature drift. The proposed approach contains a description of both, synthetic and real datasets. These datasets include different types of drifts like sudden, incremental, or gradual. Also, the location of the drift can be programmed. Additionally, we propose Java scripts that allow specifying the number and place of drift locations. The software is focused on prequential error methodology.Our proposition can be used by scientists interested in machine learning and concept drift detection because the proposed solution makes it easier to conduct practical experiments on that matter. The proposed solution allows for conducting experiments in a homogeneous programming environment. Versions of Python programs, stored in the GitHub repository, contain implementations of popular classifiers and drift detectors. The GitHub repository is located in the Public Domain.

The article presents a new concept drift detection method based on analyzing the importance of features of instances in the data stream. The data stream contains information about distribution patterns that reflect different concepts that may be hidden in the data stream. The presented drift detector concept uses information about the fluctuation of the most informative feature inside chunks of the data stream and compares it with the change of the same feature in neighbor chunks. In the case of data streams, the meaning of features can change over time. These changes affect the quality of the classification but can also be a significant indicator of ongoing concept drift. After detecting the drift, the classifier should be trained with the new dataset. But this issue is not addressed in this article.

Riskful and aggressive driving are significant problems in road transportation. Numerous studies prove drivers’ behavior is critical in most road accidents and contributes significantly to fuel consumption and emissions. In an attempt to improve road safety, it is important to be able to detect such unskilful and irresponsible drivers on a mass scale. Thus an automated onboard driving safety assessment system is required. Transmitting amounts of telemetric data required to assess driving style is unpractical and economically unjustified, thus some data processing must take place in an Onboard Unit (OBU). As a first step in driving safety assessment is detecting maneuvers, this is the task, that can be done in the OBU. In this paper, we propose a Finite State Machine based algorithm, which could run on OBU and allows marking of basic maneuvers in the telemetry datastream. This allows the calculation of additional features, aggregation, and transmission of data useful for classification the of driving safety, and thus a vast reduction in data volume sent over the air. Also, an example of detecting unsafe maneuver using such aggregated data is given.

The growing amount of collected and processed data means that there is a need to control access to these resources. Very often, this type of control is carried out on the basis of biometric analysis. The article proposes a new user authentication method based on a spatial analysis of the movement of the finger's position. This movement creates a sequence of data that is registered by a motion recording device. The presented approach combines spatial analysis of the position of all fingers at the time. The proposed method is able to use the specific, often different movements of fingers of each user. The experimental results confirm the effectiveness of the method in biometric applications. In this paper, we also introduce an effective method of feature selection, based on the Hotelling T-2 statistic. This approach allows selecting the best distinctive features of each object from a set of all objects in the database. It is possible thanks to the appropriate preparation of the input data.

Traffic accidents and vehicle mishandling are significant problems in road transportation, affecting human lives. Various studies suggest that driver behavior is a key factor in the most road accidents and contributes significantly to fuel consumption and emissions. Improvements in driver behavior can be achieved by providing feedback to drivers on their driving behavior. The identification of risky and wasteful maneuvers allows the evaluation of driver behavior. This allows the elimination of irresponsible drivers who pose a danger in traffic, and at the same time, it allows the reduction of maintenance and repair costs of the vehicle fleet. This paper presents the first stage of a driver profiling method based on the analysis of signals coming from the vehicle CAN bus and auxiliary device containing a GPS receiver and an IMU unit. No additional equipment is needed, what is an advantage of the proposed method.

This paper presents a method of personal identification via the analysis of lip print images. This remains a little explored field even within the most serious and leading biometric research teams. Biometrics is the scientific study of the identification and verification of individuals based on their physiological and behavioral traits. Such traits are permanent, unique, and can be used to separately identify any one individual from any another. In the method we present here, we integrated complex image processing techniques, machine learning, and statistical methods. We then evaluated this new method on previously collected realistic (i.e. low quality) real-world lip print images. Multi-variant experimental protocols, specifically designed for this work, then confirmed the accuracy of our new technique. Our results have extended the knowledge of – and the collection of methods available for – the identification of biometric objects from incomplete data sets. Biometric analysis techniques are rapidly gaining in importance. The approach we propose here will be useful in many areas including biometrics, forensics, and forensic medicine. The novelty of this proposed method is its ability to work on lip print images that are partially corrupted or incomplete, as often occurs in practice. Low quality areas are recognized and effectively eliminated. These areas are neither taken into account during the classifier learning process nor later during the samples’ classification. This was confirmed in a series of experiments in which the best classification accuracy achieved was 94.40%.

In this paper, a method for diagnosing Parkinson's disease based on features derived from hand-drawn spirals is presented. During drawing of these spirals on a tablet, coordinates of points of the spiral, pressure and angle of the pen at that point, and timestamp were registered. A set of features derived from the registered data, by means of which the classification was performed, has been proposed. For testing purposes, classification has been done by means of several of the most popular machine learning methods, for which the accuracy of Parkinson's disease recognition was determined. The study has proven that the proposed set of features enables the effective diagnosis of Parkinson's disease. The proposed method can be used in screening tests for Parkinson's disease. The experiments were conducted on a publicly available "Parkinson Disease Spiral Drawings Using Digitized Graphics Tablet Data Set" database from the UCI archives. This database contains drawings of spirals made by people with Parkinson's disease as well as by healthy people.

This paper describes a novel method for classification based on a partly incomplete and low quality image data. The method has been tested on images with low quality areas such as partially corrupted, dark, or blurry images. Image data are converted to classifier input. As a result, we are dealing with missing data describing these areas, but the classifier was designed to work in the presence of missing data. The analysis of the literature shows that missing data have a negative impact on the operation of the learning algorithms. This field is still insufficient explored. In our method, we integrate complex image processing techniques, machine learning, and statistical methods under various contexts. The novelty of the presented approach is the structure of the classifier, which works in the presence of missing data on different principles than those used so far. Evaluation was performed on the basis of realistic images, collected before experiments. Multi-variant experimental protocols, specially designed in this work, confirmed the accuracy of the method. The obtained results help to better understand the methods of recognizing objects with incomplete data structures. Analysis techniques proposed in this paper are rapidly gaining in importance, so the proposed approach can be useful in many areas where the machine learning is employed.

Common problem in data classification is the incompleteness of the data, and not always it is possible to re-acquire the missing values. Another approach is to fill-in missing values using some statistical methods. This however distracts the original data and may lead to over-fit the classifier to the artificially generated values, and in consequence to overestimate the classifier accuracy in Cross Validation tests. In this paper we propose a solution where, for a reference data consisting of complete and incomplete records, complete records serve as a reference data for a standard classifier, while the whole set serves as a reference data for single feature subspaced classifier.

Classifiers are used in a variety of applications, among them the classification of medical data. Their efficiency depends on the quality of training data, which is a disadvantage in the case of medical data that are often imperfect (e.g. incomplete, imbalanced, uncertain). Moreover, numerous classifiers are black-boxes from the perspective of diagnosticians who perform the final diagnoses. These drawbacks degrade the potential usefulness of classifiers in diagnosis support. A rule-based reasoning may overcome these mentioned limitations. We introduce both a rule selection and a diagnosis support method based on the Dempster–Shafer and fuzzy set theories. The theories can manage an interpretation of incomplete and imbalanced data, imprecision of medical information and knowledge uncertainty. The usefulness of the method will be proven on a test case of liver fibrosis diagnosis. The liver fibrosis stage is difficult to recognize even for experienced physicians. The diagnosis of the liver state by an invasive biopsy is ambiguous and dependent on its finite precision. Therefore, knowledge-based methods are being sought to reduce the need of invasive testing. We use a real medical database related to patients affected by hepatitis C virus to extract knowledge. The database has missing and outlying values and patients’ diagnoses are uncertain. The proposed methods provide simple diagnostic rules that are helpful in this study of liver fibrosis and in processing deficient data. The greatest benefit and novelty of the approach is the ability to assess three stages of fibrosis in a non-invasive way, whereas other medical tests allow to detect only the last stage, i.e. the cirrhosis.

This paper presents a new bootloader for Atmel's AVR ATMega microcontroller family. Presented solution compared to Optiboot is smaller and faster, leaving programmer more flash space, and reducing time required to upload, or to update firmware of the microcontroller. Concept of bootloader made the Arduino so popular, due to removing a need for specialized programming device in the prototyping stage. However, the original idea of bootloader was to enable end-user or less qualified service personnel to update firmware of the final product/embedded device.

Many datasets, especially various historical medical data are incomplete. Various qualities of data can significantly hamper medical diagnosis and are bottlenecks of medical support systems. Nowadays, such systems are often used in medical diagnosis. Even great number of data can be unsuitable when data is imbalanced, missing or corrupted. In some cases these troubles can be overcome by machine learning algorithms designed for predictive modeling.Proposed approach was tested on real medical data and some benchmarks dataset form UCI repository. The liver fibrosis disease from a medical point of view is difficult to treatment and has a significant social and economic impact. Stages of liver fibrosis are diagnosed by clinical observation and evaluations, coupled with a so-called METAVIR rating scale. However, these methods may be insufficient, especially in the recognition of phase of the disease. This paper describes a newly developed algorithm to non-invasive fibrosis stage recognition using machine learning methods - a classification model based on feature projection k-NN classifier. This solution allows extracting data characteristics from the historical data which may be incomplete and may contain imbalance (unequal) sets of patients. Proposed novel solution is based on peripheral blood analysis without using any specialized biomarkers, and can be successfully included to medical diagnosis support systems and might be a powerful tool for effective estimation of liver fibrosis stages. (C) 2016 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier Sp. z o.o. All rights reserved.