Search Results (209)

Predicting customer churn is essential for telecommunications companies to maintain profitability. However, training models on historical models leads to performance degradation when they are applied to future conditions—a phenomenon known as concept drift. We employ a sliding-window approach that separates the training and testing time windows, creating a future-based “true test”. Using unique real data, we show that a CatBoost classifier model trained on older data can remain relevant when new, unseen intervals are used. A key innovation of our work is the use of 40-day “partial churn” labels; a model trained on these labels accurately predicts 90-day churn by simply adjusting the decision threshold. Out of the six modeled scenarios, in the main realistic scenario, CatBoost retained an accuracy above 0.798 and an F1 of near 0.704, reflecting its robustness even under real-world delays and potential drift. Overall, our findings emphasize that models do not necessarily “expire” with time; rather, their performance varies according to when they are tested. This research underscores the importance of a truly future-based evaluation (instead of artificial splits) and offers practical guidance for earlier churn detection when facing real-world data delays. Full article

Supraventricular arrhythmias (SVAs), including the often-asymptomatic supraventricular extrasystole (SVE), pose significant challenges in early detection and precise diagnosis. These challenges are of paramount importance, as recurrent SVEs may elevate the risk of developing severe SVAs, potentially resulting in cardiac weakening and subsequent heart failure. In the study conducted, an innovative approach was introduced that combined a convolutional neural network (CNN) architecture to enable the early identification and characterization of SVEs within electrocardiogram (ECG) signals. The analysis leveraged a dataset comprising 78 half-hour recordings from the highly regarded MIT-BIH Arrhythmia Database, which included annotation headers serving as labels for each recording. Signals were down-sampled by a factor of 2 and split into windows of 512 samples, with 12,288 observations for training. Following the methodology, classic signal preprocessing techniques (filtering and data normalization) were used. The proposed model was based on the UNET 1D model. A binary cross-entropy loss function, Adam optimizer, and a batch size of 128 were obtained after a hyperparameter tuning. As a training-validation methodology, a 50-fold cross-validation technique was used. The approach demonstrated a Dice coefficient of 79.01%, a precision of 80.96%, and a recall rate of 86.60% in detecting SVE events. These findings were corroborated through meticulous comparison with the annotations provided by the MIT-BIH database. The results underscore the immense potential of CNN and deep learning techniques in the early detection of supraventricular arrhythmias. This approach not only offers a valuable tool for healthcare professionals engaged in telemonitoring and early intervention strategies but also represents a significant contribution to the field of cardiac health monitoring. By facilitating efficient and precise identification of SVEs, our research sets the stage for improved patient outcomes and the prevention of severe SVAs, marking substantial advancements in this critical domain. Full article

With the increasing availability of wearable devices for data collection, studies in human activity recognition have gained significant popularity. These studies report high accuracies on k-fold cross validation, which is not reflective of their generalization performance but is a result of the inappropriate split of testing and training datasets, causing these models to evaluate the same subjects that they were trained on, making them subject-dependent. This study comparatively discusses this validation approach with a universal approach, Leave-One-Subject-Out (LOSO) cross-validation which is not subject-dependent and ensures that an entirely new subject is used for evaluation in each fold, validated on four different machine learning models trained on windowed data and select hand-crafted features. The random forest model, with the highest accuracy of 76% when evaluated on LOSO, achieved an accuracy of 89% on k-fold cross-validation, demonstrating data leakage. Additionally, this experiment underscores the significance of hand-crafted features by contrasting their accuracy with that of raw sensor models. The feature models demonstrate a remarkable 30% higher accuracy, underscoring the importance of feature engineering in enhancing the robustness and precision of HAR systems. Full article

Hybrid production plants harness diverse climatic sources for electricity generation, playing a crucial role in the transition to renewable energies. This study aims to forecast the profitability of a combined wind–photovoltaic energy system. Here, we develop a model that integrates predicted spot prices and electricity output forecasts, incorporating relevant climatic variables to enhance accuracy. The jointly modeled climatic variables and the spot price constitute one of the innovative aspects of this work. Regarding practical application, we considered a hypothetical wind–photovoltaic plant located in Italy and used the relevant climate series to determine the quantity of energy produced. We forecast the quantity of energy as well as income through machine learning techniques and more traditional statistical and econometric models. We evaluate the results by splitting the dataset into estimation windows and test windows, and using a backtesting technique. In particular, we found evidence that ML regression techniques outperform results obtained with traditional econometric models. Regarding the models used to achieve this goal, the objective is not to propose original models but to verify the effectiveness of the most recent machine learning models for this important application, and to compare them with more classic linear regression techniques. Full article

Recently, a new set of biometric traits, called medical biometrics, have been explored for human identity verification. This study introduces a novel framework for recognizing human identity through heart sound signals, commonly referred to as phonocardiograms (PCGs). The framework is built on extracting and suitably processing Mel-Frequency Cepstral Coefficients (MFCCs) from PCGs and on a classifier based on a Multilayer Perceptron (MLP) network. A large dataset containing heart sounds acquired from 206 people has been used to perform the experiments. The classifier was tuned to obtain the same false positive and false negative misclassification rates (equal error rate: EER = FPR = FNR) on chunks of audio lasting 2 s. This target has been reached, splitting the dataset into 70% and 30% training and testing non-overlapped subsets, respectively. A recurrence filter has been applied to also improve the performance of the system in the presence of noisy recordings. After the application of the filter on chunks of audio signal lasting from 2 to 22 s, the performance of the system has been evaluated in terms of recall, specificity, precision, negative predictive value, accuracy, and F1-score. All the performance metrics are higher than 97.86% with the recurrence filter applied on a window lasting 22 s and in different noise conditions. Full article

Remote sensing satellite data have been a crucial tool in understanding urban climates. The variety of sensors with different spatiotemporal characteristics and retrieval methodologies gave rise to a multitude of approaches when analyzing the surface urban heat island effect (SUHI). Although there are considerable advantages that arise from these different characteristics (spatiotemporal resolution, time of observation, etc.), it also means that there is a need for understanding the ability of sensors in capturing spatial and temporal SUHI patterns. For this, several land surface temperature products are compared for the cities of Madrid and Paris, retrieved from five sensors: the Spinning Enhanced Visible and InfraRed Imager onboard Meteosat Second Generation, the Advanced Very-High-Resolution Radiometer onboard Metop, the Moderate-resolution Imaging Spectroradiometer onboard both Aqua and Terra, and the Thermal Infrared Sensor onboard Landsat 8 and 9. These products span a wide range of LST algorithms, including split-window, single-channel, and temperature–emissivity separation methods. Results show that the diurnal amplitude of SUHI may not be well represented when considering daytime and nighttime polar orbiting platforms. Also, significant differences arise in SUHI intensity and spatial and temporal variability due to the different methods implemented for LST retrieval. Full article

Land surface temperature (LST) is important in a variety of applications, such as urban thermal environment monitoring and water resource management. In this paper, eleven candidate split-window (SW) algorithms were adapted to Thermal Infrared Sensor-2 (TIRS-2) data of the Landsat 9 satellite for estimating the LST. The simulated dataset produced by extensive radiative transfer modeling and five global atmospheric profile databases was used to determine the SW algorithm coefficients. Ground measurements gathered at Surface Radiation Budget Network sites were used to confirm the efficiency of the SW algorithms after their performance was initially examined using the independent simulation dataset. Five atmospheric profile databases perform similarly in training accuracy under various subranges of total water vapor. The candidate SW algorithms demonstrate superior performance compared to the radiative transfer equation algorithm, exhibiting a reduction in overall bias and RMSE by 1.30 K and 1.0 K, respectively. It is expected to provide guidance for the generation of the Landsat 9 LST using the SW algorithms. Full article

This paper presents a fault diagnosis method for a vacuum contactor using the generalized Stockwell transform (GST) of vibration signals. The objective is to solve the problem of low diagnostic performance efficiency caused by the inadequate feature extraction capability and the redundant pixels in the graph background. The proposed method is based on the time-frequency graph optimization technique and ShuffleNetV2 network. Firstly, vibration signals in different states are collected and converted into GST time-frequency graphs. Secondly, multi-resolution GST time-frequency graphs are generated to cover signal characteristics in all frequency bands by adjusting the GST Gaussian window width factor λ. The OTSU algorithm is then combined to crop the energy concentration area, and the size of these time-frequency graphs is optimized by 68.86%. Finally, considering the advantages of the channel split and channel shuffle methods, the ShuffleNetV2 network is adopted to improve the feature learning ability and identify fault categories. In this paper, the CKJ5-400/1140 vacuum contactor is taken as the test object. The fault recognition accuracy reaches 99.74%, and the single iteration time of model training is reduced by 19.42%. Full article

Several techniques have been described for maxillary sinus graft augmentation, including the lateral window technique and crestal approach with osteotomes or osseodensification. Platelet-rich fibrin has been used in maxillary sinus lift procedures due to its ability to accelerate soft and hard tissue healing. The aim of this study was to evaluate the potential of PRF in combination with the synthetic hydroxyapatite NanoBone^® to enhance bone regeneration in sinus floor elevation with the lateral window technique. Out of the 50 individuals screened in a preoperative assessment visit from the CESPU—Famalicão clinical unit and intervened upon between January 2023 and December 2023, only 6 patients who met the study’s inclusion criteria consented to participate. In a split-mouth study, twelve sinus graft surgeries were carried out. Our observations reveal that for the test group (NanoBone^®/PRF), there is a 27.5 ± 4.9% increase new vital bone, 23.0 ± 3.7% increase in inert bone particles, and 49.4 ± 2.8% increase in connective tissue. Meanwhile, for the control group (NanoBone^®), there is a 19.5 ± 3.0% increase in new vital bone, 23.4 ± 5.7% increase in inert bone particles, and 57.0 ± 3.5% increase in connective tissue. The results strongly indicate that mixing liquid PRF with NanoBone^® does not have a negative influence on the amount of viable bone formation, and it seems to slightly increase the amount of new bone formation and revascularization in sinus bone graft procedures with the lateral window technique compared to the single use of NanoBone^®. Full article

The NOAA-AVHRR (National Oceanographic and Atmospheric Administration–Advanced Very High-Resolution Radiometer) archive includes data from 1981 onwards, which allow for estimating land surface temperature (LST), a key parameter for the study of global warming as well as surface characterization. However, algorithms for LST retrieval were developed before the latest sensors and were based on more reduced atmospheric datasets. Here, we present 50 novel sets of coefficients for an LST retrieval algorithm from NOAA-AVHRR sensors, to which we added one historical methodology, which we validate against historical in situ as well as independent satellite data. This validation shows that the historical algorithm performs surprisingly well, with an in situ RMSE below 1.5 K and a quasi-null bias when compared with independent satellite data. A couple of the novel algorithms also perform within expectations (errors below 1.5 K), so any of these could be used for the complete processing of the AVHRR dataset. In our case, considering consistency with previous works, we opt for the use of the historical algorithm, now also tested for more recent periods. Full article

Accurate land surface temperature (LST) retrieval from satellite data is pivotal in environmental monitoring and scientific research. This study addresses the impact of variability in the effective wavelengths used for LST retrieval from the Thermal Infrared Sensor (TIRS) data of Landsat 8. We conduct a detailed analysis comparing the effective wavelengths reported by Yu et al. (2014) and those derived from data provided by the USGS. Our analysis reveals significant variability in the effective wavelengths for bands 10 and 11 of Landsat 8. By applying Planck’s Law and utilizing the K1 and K2 coefficients available in the metadata of Landsat 8 products, we derive the effective wavelengths for bands 10 and 11. We also rederive the effective wavelength by integrating the spectral response function of the TIRS1 sensor. Our findings indicate that the effective wavelength for band 10 is 10.814 μm, aligning with the USGS data, while the effective wavelength for band 11 is 12.013 μm. We discuss the implications of these corrected effective wavelengths on the accuracy of LST retrieval algorithms, particularly the single channel algorithm (SC) and the radiative transfer equation (RT) employed by Yu et al. The importance of using precise effective wavelengths in satellite-based temperature retrieval is emphasized, to ensure the reliability and consistency of results. This analysis underscores the critical role of accurate spectral calibration parameters in remote sensing studies and provides valuable insights in the field of land surface temperature retrieval from Landsat 8 TIRS data. Full article

Changes in land use and land cover (LULC) have a significant impact on urban planning and environmental dynamics, especially in regions experiencing rapid urbanization. In this context, by leveraging the Google Earth Engine (GEE), this study evaluates the effects of land use and land cover modifications on surface temperature in a semi-arid zone of northwestern Algeria between 1989 and 2019. Through the analysis of Landsat images on GEE, indices such as normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), and normalized difference latent heat index (NDLI) were extracted, and the random forest and split window algorithms were used for supervised classification and surface temperature estimation. The multi-index approach combining the Normalized Difference Tillage Index (NDTI), NDBI, and NDVI resulted in kappa coefficients ranging from 0.96 to 0.98. The spatial and temporal analysis of surface temperature revealed an increase of 4 to 6 degrees across the four classes (urban, barren land, vegetation, and forest). The Google Earth Engine approach facilitated detailed spatial and temporal analysis, aiding in understanding surface temperature evolution at various scales. This ability to conduct large-scale and long-term analysis is essential for understanding trends and impacts of land use changes at regional and global levels. Full article

When a flood strikes, the two most critical tasks are evacuation and relief distribution. It is essential to integrate these tasks, particularly before the floodwater reaches the vulnerable area, to minimize loss and damage. This paper presents a mathematical model of vehicle routing problems to optimize an integrated disaster relief operation. The model addresses routing for both the evacuation and relief distribution tasks in the early stages of a flood, aiming to identify a minimal number of vehicles required with their corresponding routes to transport vulnerable individuals and simultaneously distribute emergency relief. The new model incorporates several features, including vehicle reuse, multi-trip and split delivery scenarios for evacuees and emergency relief items, uncertainty in evacuation demands, and closing time windows at evacuation points. Due to the complexity of vehicle routing problems, particularly in large-scale scenarios, the exact approach for obtaining optimal solutions is time-consuming. Therefore, we propose the use of a metaheuristic algorithm, specifically a modified genetic algorithm, to find an approximate solution for the proposed model. We apply the developed model and modified algorithm to various simulated flood scenarios and a real-life case study from Indonesia. The experimental results demonstrate that our approach requires fewer vehicles compared to standard models for similar scenarios. Moreover, while the exact approach fails to find optimal solutions within a reasonable timeframe for large-scale scenarios, our new approach provides near-optimal solutions in a much shorter time. In smaller simulated scenarios, the modified genetic algorithm obtains optimal or near-optimal solutions approximately 92.5% faster than the exact approach. Full article

In the last few years, many space surveillance initiatives have started to consider the problem represented by resident space object overpopulation. In particular, the European Space Surveillance and Tracking (EUSST) consortium is in charge of providing services like collision avoidance, fragmentation analysis, and re-entry, which rely on measurements obtained through ground-based sensors. BIRALES is an Italian survey radar belonging to the EUSST framework and is capable of providing measurements including Doppler shift, slant range, and angular profile. In recent years, the Music Approach for Track Estimate and Refinement (MATER) algorithm has been developed to retrieve angular tracks through an adaptive beamforming technique, guaranteeing the generation of more accurate and robust measurements with respect to the previous static beamforming approach. This work presents the design of a new data processing chain to be used by BIRALES to compute the angular track. The signal acquired by the BIRALES receiver array is down-converted and the receiver bandwidth is split into multiple channels, in order to maximize the signal-to-noise ratio of the measurements. Then, the signal passes through a detection block, where an isolation procedure creates, for each epoch, signal correlation matrices (CMs) related to the channels involved in the detection and then processes them to isolate the data stream related to a single detected source. Consequently, for each epoch and for each detected source, just the CM featuring the largest signal contribution is kept, allowing deriving the Doppler shift measurement from the channel illumination sequence. The MATER algorithm is applied to each CM stream, first estimating the signal directions of arrival, then grouping them in the observation time window, and eventually returning the target angular track. Ambiguous estimates may be present due to the configuration of the receiver array, which cause spatial aliasing phenomena. This problem can be addressed by either exploiting transit prediction (in the case of cataloged objects), or by applying tailored criteria (for uncatalogued objects). The performance of the new architecture was assessed in real operational scenarios, demonstrating the enhancement represented by the implementation of the channelization strategy, as well as the angular measurement accuracy returned by MATER, in both nominal and off-nominal scenarios. Full article

Streamflow prediction is crucial for planning future developments and safety measures along river basins, especially in the face of changing climate patterns. In this study, we utilized monthly streamflow data from the United States Bureau of Reclamation and meteorological data (snow water equivalent, temperature, and precipitation) from the various weather monitoring stations of the Snow Telemetry Network within the Upper Colorado River Basin to forecast monthly streamflow at Lees Ferry, a specific location along the Colorado River in the basin. Four machine learning models—Random Forest Regression, Long short-term memory, Gated Recurrent Unit, and Seasonal AutoRegresive Integrated Moving Average—were trained using 30 years of monthly data (1991–2020), split into 80% for training (1991–2014) and 20% for testing (2015–2020). Initially, only historical streamflow data were used for predictions, followed by including meteorological factors to assess their impact on streamflow. Subsequently, sequence analysis was conducted to explore various input-output sequence window combinations. We then evaluated the influence of each factor on streamflow by testing all possible combinations to identify the optimal feature combination for prediction. Our results indicate that the Random Forest Regression model consistently outperformed others, especially after integrating all meteorological factors with historical streamflow data. The best performance was achieved with a 24-month look-back period to predict 12 months of streamflow, yielding a Root Mean Square Error of 2.25 and R-squared ($R^2$) of 0.80. Finally, to assess model generalizability, we tested the best model at other locations—Greenwood Springs (Colorado River), Maybell (Yampa River), and Archuleta (San Juan) in the basin. Full article