Stacking Technique Research Articles

Stacking-based modelling for improved over-indebtedness predictions

In a world now starkly divided into pre- and post-COVID times, it's imperative to examine the impact of this public health crisis on the banking functions - particularly over-indebtedness risks. In this work, a flexible analytics-based model is proposed to improve the banking process of detecting customers who are likely to have difficulty in managing their debt. The proposed model assists the banks in improving their predictions. The proposed meta-model extracts information from existing data to determine patterns and to predict future outcomes and trends. We test and evaluate a large variety of Machine Learning Algorithms (MLAs) by using new techniques like feature selection. Moreover, models of previous months are combined in order to build a meta-model representing several months using stacked generalisation technique. The new model will identify 91% of the customers potentially unable to repay their debt six months ahead and enable the bank to implement targeted collections strategies.

Multilayer Stacked Ensemble Learning Model to Detect Phishing Websites

Phishing is a cyber attack that tricks the online users into revealing sensitive information with a fake website imitating a legitimate website. The attackers with stolen credentials not only use them for the targeted website but also can be used for accessing the other popular legitimate websites. There exists many anti-phishing techniques, toolbars, extensions to counter the phishing sites but still the phishing attacks are major concern in the current digital world. In this paper, we propose a multilayered stacked ensemble learning technique which consists of estimators at different layers where the predictions of estimators from current layer are fed as input to the next layer. From the experimental results, it is observed that the proposed model achieved a significant performance when evaluated with different datasets with an accuracy of ranging from 96.79%to 98.90%. The proposed model is evaluated with datasets from UCI(D1), Mendeley 2018(D2) and Mendeley 2020(D3,D4). The proposed model achieved detection rate of 97.76% with D1 dataset, achieved an accuracy of 98.9% with D2 dataset. Finally, the technique is tested with D3 and D4 which resulted in accuracy of 96.79% and 98.43% respectively. Also, the proposed model outperformed baseline models corresponding to datasets with a significant difference in accuracy and F score metrics.

Evaluation quantitative structure-activity relationship (QSAR) using ensemble learning methods on acetylcholinesterase inhibitors for Alzheimer's disease

Acetylcholinesterase inhibitors (AChEI) are among the most potential drug molecules for treating Alzheimer's disease and effectively treating its symptoms. Quantitative Structure and Activity Relationship (QSAR) is a computational modeling method to determine the relationship between the structural properties of chemical compounds and biological activities. This study used a classification QSAR model to predict the active and inactive molecules in AChEI. There were 3809 molecules of compounds in the preprocessing stage consisting of 2215 molecules of active compounds and 1594 molecules of inactive compounds. The compound molecules in SMILES were extracted into the fingerprint using the ECFP and FCFP method with diameters of 4 and 6. In this study, the ensemble learning methods used to build the classification QSAR model were voting, averaging, and stacking. The results showed that the ensemble learning method had a better performance than using only one base model. The classification QSAR model with base model obtained an accuracy of 92%, a sensitivity of 89.97%, a specificity of 93%, and an MCC of 83%. The comparison, the ensemble learning method with the stacking technique obtained an accuracy of 93%, a sensitivity of 92%, a specificity of 94%, and an MCC of 86%.

Predicting Mental Health Illness using Machine Learning Algorithms

Early detection of mental health issues allows specialists to treat them more effectively and it improves patient’s quality of life. Mental health is about one’s psychological, emotional, and social well-being. It affects the way how one thinks, feels, and acts. Mental health is very important at every stage of life, from childhood and adolescence through adulthood. This study identified five machine learning techniques and assessed their accuracy in identifying mental health issues using several accuracy criteria. The five machine learning techniques are Logistic Regression, K-NN Classifier, Decision Tree Classifier, Random Forest, and Stacking. We have compared these techniques and implemented them and also obtained the most accurate one in Stacking technique based with an accuracy of prediction 81.75%.

A Compact CMOS Broadband Bidirectional Digital Transceiver Frontend With Capacitor Bank and Transformer Matching Network Reuse

This article presents a fully integrated bidirectional class-G digital Doherty switched capacitor transmitter (TX) and N-path Quadrature receiver (RX) in CMOS. Through sharing on-chip capacitor banks, typically occupying a major portion of the digital TX or RX chip area, as well as the RF passive matching networks, the overall size can be radically reduced. Moreover, the overall performance could be further improved by eliminating the need for an integrated T/RX switch and its corresponding loss and area overhead. The class-G operation is used within the Doherty TX to increase the output power and backoff efficiency, while the capacitive stacking technique is used in the RX to increase the voltage gain. A transformer network is used to present the optimum impedance for both the parallel Doherty TX and RX mode of operation, as well as the class-G Doherty active load modulation. As a proof-of-concept, the joint bidirectional class-G digital Doherty switched-capacitor TX and N-path Quadrature RX through capacitor bank sharing is implemented in a 45-nm CMOS SOI process. The TX demonstrates a P <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">out</sub> 1dB bandwidth (BW) of 1.6-3.1 GHz, a fractional BW >63%, and peak output power (P <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">out</sub> ) of 22.5dBm at 2.4GHz. The peak drain efficiency (DE) of the TX is 49.5% at 1.8GHz and 41.5%/38.7%/31.6%/18.1% for the peak/2.5/6/12dB power back off (PBO) at 2.4GHz. The DE improvement compared to class-B PA is 1.24×/1.51×/1.72× at 2.5/6/12dB PBO. The TX is measured using 64-QAM/20MHz modulation without the use of AM-PM pre-distortion or pattern based DPD. It achieves an excellent -27.1dB EVM, -31.31dBc ACLR, 14.6dBm average P <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">out</sub> and 25.8% average DE at 1.6GHz. The RX achieves a noise figure (NF) of 7.6dB at 2.2GHz and a conversion gain of 17dB with a 12 MHz bandwidth. In addition, the proposed RX front-end achieves < -60 dBm LO leakage over the operating frequency range.

Two-station analysis of passive surface waves with continuous wavelet transform and plane-wave-based beamforming

Using the cross-correlation or cross-coherence method to extract the traveltime information of surface waves between two stations from the ambient noise seismic data has become a popular method, which is widely utilized in shallow, regional, and global-scale geophysical surveys. Different from the common-used narrow-band filter (NBF), we adopt the continuous wavelet transform (CWT) as an alternative method to measure the phase velocity from the interstation cross-correlation function (CCF). On account of the distribution of urban noisy sources it is hard to satisfy the requirement of random, stable, uncorrelated, and uniform, which usually makes the measured phase-velocity uncertain. For those surveys with an ultra-short recording time of noise, the adverse effect would be more severe. Hence, we propose to adopt the plane-wave-based beamforming (PWBB) to scan the incoming direction of surface waves based on the hypothesis of plane-wave propagation. To suppress the incoherent components during the stack of CCFs, the phase-weighted stack (PWS) technique is also included in the proposed workflow. Several synthetic tests have verified the good performance of the new method. Especially when only a few seismic sensors are arranged in the research area, the advantages of our method would be more obvious. Then, the workflow is applied to the analysis of the field noise data, which were collected at Fengman hydropower station, Jilin province, China. The beamforming result indicates that the main component of the signals is produced by the effects of river scouring under the reservoir gate. Finally, the S-wave velocity tomogram of the research area and its geological interpretation are given. All data tests demonstrate that the proposed CWT with PWBB can achieve the stable dispersion measurement from the recorded seismic data of two station.

Improvement of Wheat Grain Yield Prediction Model Performance Based on Stacking Technique

Crop growth and development is a dynamic and complex process, and the essence of yield formation is the continuous accumulation of photosynthetic products from multiple fertility stages. In this study, a new stacking method for integrating multiple growth stages information was proposed to improve the performance of the winter wheat grain yield (GY) prediction model. For this purpose, crop canopy hyperspectral reflectance and leaf area index (LAI) data were obtained at the jointing, flagging, anthesis and grain filling stages. In this case, 15 vegetation indices and LAI were used as input features of the elastic network to construct GY prediction models for single growth stage. Based on Stacking technique, the GY prediction results of four single growth stages were integrated to construct the ensemble learning framework. The results showed that vegetation indices coupled LAI could effectively overcome the spectral saturation phenomenon, the validated R2 of each growth stage was improved by 10%, 22.5%, 3.6% and 10%, respectively. The stacking method provided more stable information with higher prediction accuracy than the individual fertility results (R2 = 0.74), and the R2 of the model validation phase improved by 236%, 51%, 27.6%, and 12.1%, respectively. The study can provide a reference for GY prediction of other crops.

Modeling and experimental validation of a quantitative bar-type corrosion measuring probe using piezoelectric stack and electromechanical impedance technique

Corrosion coupon method has been widely used to estimate the corrosion rate in multiple industries. However, in this method, the weights of the coupons are measured periodically, which limit the application for on-line monitoring. In this paper, a novel type of quantitative bar-type corrosion measuring probe using piezoelectric stack and electromechanical impedance (EMI) technique was proposed. The probe consists of a piezoelectric stack and a metal bar. The multilayer models were used to derive the solution of the probe in longitudinal vibration mode. Five probe prototypes with designated probe length were fabricated to simulate uniform corrosion induced mass loss and investigate the EMI response with probe length. The relationship between the corrosion induced probe length loss and the first and second resonant and anti-resonant frequencies were analyzed. The measured results agreed well with the theoretical predictions. In addition, the accelerated corrosion tests were also performed to induce corrosion to the probe in a realistic setup and further validate the efficacy of the proposed method. The present study proved the feasibility of using the proposed bar-type corrosion measuring probe to quantitatively assess the corrosion amount by introducing the longitudinal vibration with piezoelectric stack and EMI.

Heart Disease Classification Based on Hybrid Ensemble Stacking Technique

Heart Disease Classification Based on Hybrid Ensemble Stacking Technique

Accurate and explainable machine learning for the power factors of diamond-like thermoelectric materials

The application of machine learning (ML)-based methods to the study of thermoelectric (TE) materials is promising. Although conventional ML algorithms can achieve high prediction performance, their lack of interpretability severely obstructs researchers from extracting material-oriented insights from ML models. In this work, high ML-based prediction performance was achieved with respect to TE power factors (PFs), and the results were well understood by the SHapley Additive exPlanations (SHAP), a method to identify the correlations between targets and descriptors. We designed a robust PF prediction model for diamond-like compounds via a stacking technique, and the model achieved a coefficient of determination value above 0.95 on the test set. From the SHAP analysis, the PFs were negatively correlated with electronegativity and positively correlated with the descriptor “volume per atom” based on the previously reported dataset. TE domain knowledge was adopted to understand these correlations. This work shows that ML models can achieve high accuracy while exhibiting good interpretability, making them useful for materials scientists.

Data‐driven flight time prediction for arrival aircraft within the terminal area

Predicting the arrival aircraft's flight time plays a critical role in effectively optimizing and scheduling spatial-temporal resources in the terminal airspace. This paper focuses on a data-driven method for predicting the arrival flight time. First, based on the existing research, a feature set is constructed from four aspects: initial state, arrival pressure, sequencing pressure, and wind information, which are believed to affect arrival flight time significantly. Second, eight widely used models are developed to predict flight time, including linear regression models, nonlinear regression models, and tree-based ensemble models. Furthermore, the stacking technique is adopted to improve the prediction performance. Finally, take Guangzhou Baiyun International Airport as a study case to verify the proposed method's effectiveness. The results indicate that the arrival pressure (describing the arrival traffic demand) and the sequencing pressure (sketching the arrival traffic distribution) could effectively improve the prediction accuracy. The mean absolute percentage error of the predicted flight time via ATAGA and IGONO can be increased by 1%. Besides, the proposed method of extracting wind data could also improve the prediction performance. The mean absolute error of the predicted flight time via GYA can be reduced by 4.85 s.

Crustal structure and deformation in southeastern China revealed by receiver functions

• A V-shaped zone with high Vp/Vs ratio and thin crust observed by receiver functions. • Most stations located near the V-shaped zone exhibit significant crustal anisotropy. • Hot fingers grow from the mantle corner flow and result in Moho uplift in SE China. • Mantle corner flow might be a main reason for crustal thinning in the V-shaped zone. The subduction of the Paleo-Pacific and Philippine plates generates large-scale magmatism and crustal thinning in southeastern (SE) China. However, the dynamic process for crustal thinning in this region is still not well understood. In this study, a large number of receiver function data were used to investigate the crustal thickness and Vp/Vs ratio beneath SE China by the H-κ stacking technique. We also apply a comprehensive analysis method on the Moho Ps converted wave in radial and transverse receiver functions to calculate crustal seismic anisotropy with fast polarization direction and splitting time. The measured results show that the high Vp/Vs ratio (∼1.85) was mainly spread around two areas, forming a “V-shaped” zone. The western branch of the “V-shaped” zone was NS-oriented with a thin crustal thickness of ∼29.1 km beneath the Nanxiang-Jianghan basin. The eastern branch extended along the Tanlu fault with a thin crustal thickness of ∼28.8 km. Moreover, the measured crustal seismic anisotropy also shows that most stations located near the “V-shaped” zone exhibit significant crustal anisotropy with splitting times of 0.20–0.72 s. The fast directions of these stations are parallel to the direction of the V-shaped extension. Combining the geological data, we propose that the “hot fingers” may grow from mantle corner flow in the southern part of the V-shaped zone associated with the collision of the Pacific and Philippine plates. We deduce that the V-shaped crustal thinning and Moho uplift in SE China might be attributed to this mantle corner flow.

A Data-Independent Radio Frequency Fingerprint Extraction Scheme

Radio frequency fingerprint (RFF) has been utilized to mitigate spoofing attacks in open wireless environments, making use of the inherent characteristics of hardware. However, most existing RFF technologies are data-dependent, e.g., based on preambles or synchronization sequences. In this letter, we propose a novel data-independent RFF extraction scheme, called Least mean square-based Adaptive Filter and Stacking, abbreviated as LAFS, that is implemented on random data segments, like communication data. Intuitively, we extract converged tap coefficients as RFF by minimizing the divergence between the desired signal and the demodulated reference signal. To further improve the effect, we introduce a tap coefficient stacking (TSC) technique to stabilize the RFF. Our experiment on ZigBee devices shows that the proposed LAFS method successfully identifies transmitters with 98.9% accuracy at 10 dB by stacking 25 times.

Analysis On Power Gating Circuits Based Low Power VLSI Circuits (BCD Adder)

Currently, energy consumption in the digital circuit is a key design parameter for emerging mobile products. The principal cause of the power dissipation during idle mode is leakage currents, which are rising dramatically. Sub-threshold leakage is increased by the scaling of threshold voltage when gate current leakage increases because oxide thickness is scaled. With rising demands for mobile devices, leakage energy consumption has received even greater attention. Since a mobile device spends most of its time in standby mode, leakage power savings need to prolong the battery life. That is why low power has become a significant factor in CMOS circuit design. The required design and simulation of an AND gate with the BSIM4 MOS parameter model at 27 0C, supply voltage of 0,70V with CMOS technology of 65nm are the validation of the suitability of the proposed circuit technology. AND simulation. The performance parameters for the two AND input gate are compared with the current MTCMOS and SCCMOS techniques, such as sub-threshold leakage power dissipations in active and standby modes, the dynamic dissipation, and propagation period. The proposed hybrid super cutoff complete stack technique compared to the current MTCMOS technology shows a reduction in sub-threshold dissipation power dissipation by 3. 50x and 1.15x in standby modes and active modes respectively. The hybrid surface-cutting technique also shows savings of 2,50 and 1,04 in power dissipation at the sub-threshold in standby modes and active modes compared with the existing SCCMOS Technique.

Electroosmotic flow assisted pseudophase to pseudophase microextraction for stacking in capillary zone electrophoresis

A stacking technique is proposed to improve the poor detection sensitivity of capillary zone electrophoresis (CZE) with UV detection. A long injection (e.g., 12.4 cm plug) of model anionic analytes prepared in a dilute solution of hexadecyltrimethylammonium bromide (CTAB) was enriched 26–34-x (compared to a typical or 2.1 mm sample injection) via the injection of a micellar solution of sodium dodecyl sulfate (SDS) prior to CZE separation. During sample injection, the CTAB formed a stationary pseudophase coating, which trapped the analytes at the inner walls of a fused silica capillary. The SDS micelles then released the CTAB admicelles via the formation of solution CTAB-SDS catanionic micelles during SDS plug injection and voltage application. As the SDS micelles moved through the sample zone, the formation of the catanionic micelles then released and accumulated the analytes at the front of the injected SDS zone. The stacking technique is called electroosmotic flow (EOF) assisted pseudophase to pseudophase microextraction because the EOF was essential for the formation of CTAB-SDS catanionic micelles for microextraction. Also, the CTAB and SDS aggregates are both pseudophases, which were used to retain and release the analytes from the capillary wall, respectively.

Pseudophase microextraction for in-line sample concentration in micellar electrokinetic chromatography

Pseudophase microextraction (PPME) as a simple in-line sample concentration technique in micellar electrokinetic chromatography (MEKC) is presented. In contrast to popular electric-field driven stacking techniques in MEKC such as sweeping, PPME is pressure-driven. The technique afforded up to 403-2968x improvements in peak heights for fenoprop, dichlorprop, 1- and 2-naphthol compared to typical injection. Under the same MEKC conditions, the improvements in PPME were up to 23-59x better compared to sweeping. Briefly in PPME, the entire capillary was loaded (up to 20 capillary volumes) with the sample prepared in a dilute solution of cetyltrimethylammonium bromide ([CTAB] > critical surface aggregation concentration). The CTAB formed aggregates at the inner capillary walls and these aggregates acted as a stationary chromatographic pseudophase. After clean-up via flushing the capillary with purified water, the MEKC background solution (BGS) with sodium dodecyl sulfate was then introduced by pressure from the outlet end to elute the retained analytes. The analytes concentrate at front of the BGS and the front was moved to the inlet end of the capillary prior to MEKC. Optimization strategies and current limitations in PPME-MEKC are described. The linear ranges using a 4 capillary volume sample load obtained for fenoprop, dichlorprop, 1- and 2-naphthol were between 1 and 160 ng/mL (r2s ≥ 0.996), LOQs = 1–2.5 ng/mL and repeatability %RSDs (n = 6) were ≤5% (intra-day) and ≤7% (inter-day) (using low analyte concentrations 1-5x LOQ). PPME-MEKC with simple dilution of fortified real samples (no off-line sample concentration) was also able to detect low levels of dichlorprop (10 ng/mL, limit set in Australia) and 1- and 2−naphthol (7.5–15 ng/mL) in a drinking water and natural water sample, respectively (% recovery = 84–108%). The concept of PPME may find use in other modes of capillary electrophoresis and other nano-microscale separations.

Improving power theft detection using efficient clustering and ensemble classification

One of the main concerns of power generation systems around the world is power theft. This research proposes a framework that merges clustering and classification together in order to power theft detection. Due to the fact that most datasets do not have abnormal samples or are few, we have added abnormal samples to the original datasets using artificial attacks to create balance in the datasets and increase the correct detection rate. We improved the crow search algorithm (CSA) and used the weight feature of Crows to improve performance of clustering phase. Also, to create balance between diversification and intensification, we calculated the awareness probability parameter (AP) dynamically at iterations of the algorithm. To evaluate the performance, we used the cross validation technique have used the stacking technique in its training phase. The results of extensive experiments on three reference datasets showed high performance to detect power theft. The evaluation results showed that if the data is collected correctly and sufficiently, this framework can effectively detect power theft in any actual power grid. Also, for new attacks, if their patterns can be detected from the data, it is easily possible to implement these types of attacks.

Crustal thickness estimation in Indonesia using receiver function method

The existence of seismic wave velocity difference in the Earth crust and mantle creates the possibility to use earthquake data for estimating the crustal thickness utilizing the Ps conversion phase in the boundary. The radial component signal was deconvolved from the vertical component in the frequency domain to estimate receiver function for Indonesia region. We implemented the water level deconvolution techniques with a Gaussian filter of 2.5 Hz to eliminate the high frequency noise in the receiver function. The H-k stacking technique was performed to all receiver functions from each event to predict the crustal thickness and the Vp/Vs ratio below the stations. We analyzed ten azimuthally distributed teleseismic earthquakes recorded by 108 stations of BMKG. The result shows that the crustal thickness in Indonesia varies from 20 to 39.9 km. The western part of Sumatera, northern part of Sulawesi Island, and North Maluku region show generally thinner crust with value about 20 to 25 km. The North Sumatera, Central Java, and East Java show a considerably thicker crust of up to 36 km. Furthermore, our result also reveals a difference of crustal thickness about 5 km with the previous studies.

MeerKAT 21-cm H I imaging of Abell 2626 and beyond

Context. The morphology-density relation manifests the environmental dependence of the formation and evolution of galaxies as they continuously migrate through the cosmic web to ever denser environments. As gas-rich galaxies traverse the outskirts and inner regions of galaxy clusters, they experience sudden and radical changes in their gas content and star formation activity. Aims. The goal of this work is to gain an H I perspective on gas depletion mechanisms acting on galaxies and galaxy groups that are being accreted by a moderately massive galaxy cluster. We aim to study the relative importance and efficiency of processes such as ram-pressure stripping and tidal interactions as well as their dependency on the local and global environment of galaxies in the cluster core and in its surroundings. Methods. We have conducted a blind radio continuum and H I spectral line imaging survey with the MeerKAT radio telescope of a 2° × 2° area centred on the galaxy cluster Abell 2626. We have used the CARAcal pipeline to reduce the data, SoFiA to detect sources within the H I data cube, and GIPSY to construct spatially resolved information on the H I morphologies and kinematics of the H I detected galaxies. Results. We have detected H I in 219 galaxies with optical counterparts within the entire surveyed volume. We present the H I properties of each of the detected galaxies as a data catalogue and as an atlas page for each galaxy, including H I column-density maps, velocity fields, position-velocity diagrams, and global H I profiles. These data will also be used for case studies of identified ‘jellyfish’ galaxies and galaxy population studies by means of morphological classification of the direct H I detections as well as using the H I stacking technique.

Energy-Efficient Online Path Planning of Multiple Drones Using Reinforcement Learning

Drones, typically unmanned aerial vehicles (UAVs), have many purposes. However, simultaneous operation of multiple drones is challenging, considering the real-time interactions and the environment; the drone must avoid collision with the other drones or obstacles. The proposed Advanced TD3 model performs energy-efficient path planning at the edge-level drone. We modify the twin-delayed deep deterministic policy gradient (TD3), which is the state-of-the-art policy gradient reinforcement learning. The frame stacking technique considers the continuous action space of the drone to the TD3 model. During the training, we gradually increase the observation range of agents for fast and stable convergence. We train the modified TD3 model through Offline RL to reduce the overhead for the RL model training. Drones mount the converged RL model on their onboard computer. The Advanced TD3 model in the drones selects an energy-efficient path without the overhead of the training process of the RL model in real-time, considering external factors such as wind or another drone. The total energy consumption of drones in flight along with online path planning is approximately 106% of the total energy consumption of drones that follow offline path planning, even though the trained TD3 model does not require complex computations for real-time execution.