Isolation Forest Algorithm Research Articles

Determination of stress obtained from sensor-based wearable measurements using VGG16 deep learning model

<p>There are many researches carried out for different purposes on human-computer interactions. One of them is related to stress-related activity detection. Today, in people with disorders whose activity is not understood or misunderstood, the correct detection of the relevant movement can be vital in some cases. It may be more advantageous to use physiological signals in the body in determining the type of activity. Due to these important situations, two different research applications were carried out within the scope of this study in order to automatically detect four different types of stress, namely Neutral, Emotional, Mental, and Physical. For both applications, the data was first converted to images as a preprocessing. In the first stage of the research, the images of the standard dataset were presented to the VGG16 deep learning model. As a result, the highest accuracy rate was obtained as 67% for class 1 Neutral activation. In the second part of the study, an application was performed using the Isolation Forest Algorithm on the existing image data to remove outliers. The new dataset obtained were presented to the same model and detailed analyses were made. Accordingly, the maximum accuracy value was 97% in Physical activity. In the same application, the average rate for all activities was 82.5%. Briefly, the research contributes to the literature by demonstrating the significant impact of outliers on system performance through image transformations of existing time series physiological signals.</p>

Temperature Compensation of Laser Methane Sensor Based on a Large-Scale Dataset and the ISSA-BP Neural Network.

We aimed to improve the detection accuracy of laser methane sensors in expansive temperature application environments. In this paper, a large-scale dataset of the measured concentration of the sensor at different temperatures is established, and a temperature compensation model based on the ISSA-BP neural network is proposed. On the data side, a large-scale dataset of 15,810 sets of laser methane sensors with different temperatures and concentrations was established, and an Improved Isolation Forest algorithm was used to clean the large-scale data and remove the outliers in the dataset. On the modeling framework, a temperature compensation model based on the ISSA-BP neural network is proposed. The quasi-reflective learning, chameleon swarm algorithm, Lévy flight, and artificial rabbits optimization are utilized to improve the initialization of the sparrow population, explorer position, anti-predator position, and position of individual sparrows in each generation, respectively, to improve the global optimization seeking ability of the standard sparrow search algorithm. The ISSA-BP temperature compensation model far outperforms the four models, SVM, RF, BP, and PSO-BP, in model evaluation metrics such as MAE, MAPE, RMSE, and R-square for both the training and test sets. The results show that the algorithm in this paper can significantly improve the detection accuracy of the laser methane sensor under the wide temperature application environment.

NOx emission reduction of coal‐fired power plants through data‐driven model and particle swarm optimization

AbstractSeveral studies have aimed to predict and control carbon emissions from coal‐fired power plants. However, the highly complex combustion mechanisms in coal‐fired power plant boilers pose a significant challenge in direct modeling and optimization. To tackle this challenge, this study introduced a data‐driven approach along with model‐based process optimization to mitigate NOx emissions from coal‐fired power plants. The process involved collecting a 5‐month operational dataset containing 67 controllable parameters from a 500 MW coal‐fired power plant. Steady‐state data was isolated from the load output using a moving average method, followed by the application of an isolation forest algorithm to detect and remove anomalies. Correlation analysis was then used to evaluate parameter relationships and eliminate redundant ones. Subsequently, a NOx prediction model was developed, combining an extra tree regressor data‐driven prediction model with particle swarm optimization to optimize the most influential controllable parameters for reducing NOx emissions. Testing the proposed model across four different target loads consistently resulted in a reduction of over 20% in NOx emissions by optimizing boiler combustion parameters, representing a significant achievement in optimizing coal‐fired combustion.

Predicting Model of Dual-Mode Shield Tunneling Parameters in Complex Ground Using Recurrent Neural Networks and Multiple Optimization Algorithms

Based on the left tunnel of the Liuxiandong Station to Baimang Station section of Shenzhen Metro Line 13 (China), a prediction model for the advanced rate of dual-mode shield tunneling in complex strata was established to explore intelligent tunneling technology in complex ground. Firstly, geological parameters of the complex strata and on-site monitoring parameters of EPB/TBM dual-mode shield tunneling were collected, with tunneling parameters, shield tunneling mode, and strata parameters selected as input features. Subsequently, the Isolation Forest algorithm was employed to remove outliers from the original advance parameters, and an improved mean filtering algorithm was applied to eliminate data noise, resulting in the steady-state phase parameters of the shield tunneling process. The base model was chosen as the Long-Short Term Memory (LSTM) recurrent neural network. During the model training process, particle swarm optimization (PSO), genetic algorithm (GA), differential evolution (DE), and Bayesian optimization (BO) algorithms were, respectively, combined to optimize the model’s hyperparameters. Via rank analysis based on evaluation metrics, the BO-LSTM model was found to have the shortest runtime and highest accuracy. Finally, the dropout algorithm and five-fold time series cross-validation were incorporated into the BO-LSTM model, creating a multi-algorithm-optimized recurrent neural network model for predicting tunneling speed. The results indicate that (1) the Isolation Forest algorithm can conveniently identify outliers while considering the relationship between tunneling speed and other parameters; (2) the improved mean filtering algorithm exhibits better denoising effects on cutterhead speed and tunneling speed; and (3) the multi-algorithm optimized LSTM model exhibits high prediction accuracy and operational efficiency under various geological parameters and different excavation modes. The minimum Mean Absolute Percentage Error (MAPE) prediction result is 8.3%, with an average MAPE prediction result below 15%.

Hybrid computing framework security in dynamic offloading for IoT-enabled smart home system.

In the distributed computing era, cloud computing has completely changed organizational operations by facilitating simple access to resources. However, the rapid development of the IoT has led to collaborative computing, which raises scalability and security challenges. To fully realize the potential of the Internet of Things (IoT) in smart home technologies, there is still a need for strong data security solutions, which are essential in dynamic offloading in conjunction with edge, fog, and cloud computing. This research on smart home challenges covers in-depth examinations of data security, privacy, processing speed, storage capacity restrictions, and analytics inside networked IoT devices. We introduce the Trusted IoT Big Data Analytics (TIBDA) framework as a comprehensive solution to reshape smart living. Our primary focus is mitigating pervasive data security and privacy issues. TIBDA incorporates robust trust mechanisms, prioritizing data privacy and reliability for secure processing and user information confidentiality within the smart home environment. We achieve this by employing a hybrid cryptosystem that combines Elliptic Curve Cryptography (ECC), Post Quantum Cryptography (PQC), and Blockchain technology (BCT) to protect user privacy and confidentiality. Additionally, we comprehensively compared four prominent Artificial Intelligence anomaly detection algorithms (Isolation Forest, Local Outlier Factor, One-Class SVM, and Elliptic Envelope). We utilized machine learning classification algorithms (random forest, k-nearest neighbors, support vector machines, linear discriminant analysis, and quadratic discriminant analysis) for detecting malicious and non-malicious activities in smart home systems. Furthermore, the main part of the research is with the help of an artificial neural network (ANN) dynamic algorithm; the TIBDA framework designs a hybrid computing system that integrates edge, fog, and cloud architecture and efficiently supports numerous users while processing data from IoT devices in real-time. The analysis shows that TIBDA outperforms these systems significantly across various metrics. In terms of response time, TIBDA demonstrated a reduction of 10-20% compared to the other systems under varying user loads, device counts, and transaction volumes. Regarding security, TIBDA's AUC values were consistently higher by 5-15%, indicating superior protection against threats. Additionally, TIBDA exhibited the highest trustworthiness with an uptime percentage 10-12% greater than its competitors. TIBDA's Isolation Forest algorithm achieved an accuracy of 99.30%, and the random forest algorithm achieved an accuracy of 94.70%, outperforming other methods by 8-11%. Furthermore, our ANN-based offloading decision-making model achieved a validation accuracy of 99% and reduced loss to 0.11, demonstrating significant improvements in resource utilization and system performance.

UNSUPERVISED MACHINE LEARNING FOR SEISMIC ANOMALY DETECTION: ISOLATION FOREST ALGORITHM APPLICATION TO INDONESIAN EARTHQUAKE DATA

Indonesia's position on the seismically active Pacific "Ring of Fire" necessitates advanced methods for earthquake detection and preparedness. This paper introduces an application of the Isolation Forest algorithm—an unsupervised machine learning technique—for detecting seismic anomalies in Indonesia's complex geotectonic landscape. Unlike traditional methods that rely on predefined thresholds or patterns, the Isolation Forest algorithm isolates anomalies based on their rarity and distinctness without the need for labeled data. We applied this algorithm to a comprehensive dataset from the Indonesian Meteorology, Climatology, and Geophysical (BMKG) Agency, featuring a decade's worth of seismic events, to identify outliers that may signify potential seismic hazards. Our findings reveal that the Isolation Forest algorithm effectively identifies seismic anomalies, with 874 out of tens of thousands of events flagged as statistically significant outliers. A comparative analysis with traditional seismic anomaly detection models highlights the robustness of the Isolation Forest algorithm in handling the high-dimensional and noisy nature of seismic data, emphasizing its superiority in detecting subtle anomalies

Unraveling C-to-U RNA editing events from direct RNA sequencing

ABSTRACT In mammals, RNA editing events involve the conversion of adenosine (A) in inosine (I) by ADAR enzymes or the hydrolytic deamination of cytosine (C) in uracil (U) by the APOBEC family of enzymes, mostly APOBEC1. RNA editing has a plethora of biological functions, and its deregulation has been associated with various human disorders. While the large-scale detection of A-to-I is quite straightforward using the Illumina RNAseq technology, the identification of C-to-U events is a non-trivial task. This difficulty arises from the rarity of such events in eukaryotic genomes and the challenge of distinguishing them from background noise. Direct RNA sequencing by Oxford Nanopore Technology (ONT) permits the direct detection of Us on sequenced RNA reads. Surprisingly, using ONT reads from wild-type (WT) and APOBEC1-knock-out (KO) murine cell lines as well as in vitro synthesized RNA without any modification, we identified a systematic error affecting the accuracy of the Cs call, thereby leading to incorrect identifications of C-to-U events. To overcome this issue in direct RNA reads, here we introduce a novel machine learning strategy based on the isolation Forest (iForest) algorithm in which C-to-U editing events are considered as sequencing anomalies. Using in vitro synthesized and human ONT reads, our model optimizes the signal-to-noise ratio improving the detection of C-to-U editing sites with high accuracy, over 90% in all samples tested. Our results suggest that iForest, known for its rapid implementation and minimal memory requirements, is a promising tool to denoise ONT reads and reliably identify RNA modifications.

Anomaly Detection of Axial Piston Pump Based on the DTW-RCK-IF Composite Method Using Pressure Signals

Axial piston pumps are critical components of hydraulic systems due to their compact design and high volumetric efficiency, making them widely used. However, they are prone to failure in harsh environments characterized by high pressure and heavy loads over extended periods. Therefore, detecting abnormal behavior in axial piston pumps is of significant importance. Traditional detection methods often rely on vibration signals from the pump casings; however, these signals are susceptible to external environmental interference. In contrast, pressure signals exhibit greater stability. In this study, we propose a novel anomaly detection method for axial piston pumps, referred to as DTW-RCK-IF, which combines dynamic time warping (DTW) for data segmentation, a random convolutional kernel (RCK) for feature extraction, and isolation forest (IF) for anomaly detection using pressure signals. The model is trained using normal operating data to enable the effective detection of abnormal states. First, the DTW algorithm is employed to segment the raw data, ensuring a high degree of similarity between the segmented data. Next, the random convolutional kernel approach is used in a convolutional neural network for feature extraction, resulting in features that are representative of normal operating conditions. Finally, the isolation forest algorithm calculates the anomaly scores for anomaly detection. Experimental simulations on axial piston pumps demonstrate that, compared with vibration signals, the DTW-RCK-IF approach using pressure signals yields superior results in detecting abnormal data, with an average F1 score of 98.79% and a good fault warning effect. Validation using the publicly available CWRU-bearing and XJTU-SY-bearing full-life datasets further confirms the effectiveness of this method, with average F1 scores of 99.35% and 99.73%, respectively, highlighting its broad applicability and potential for widespread use.

OPM-MEG bad channel identification method based on the improved box-isolation forest algorithm

Bad channels result from the inherent instability of wearable optically pumped magnetometers (OPMs). These unreliable channels significantly impede subsequent signal analysis. The signal space separation method for OPMs, which relies on time-domain features such as the z-score, cannot accurately identify these unreliable channels. We introduce a novel approach for automatic bad channel identification in OPMs, utilizing frequency domain features. Box-plot technique enables predicting the number of unreliable channels. The predicted ratio of unreliable channels is incorporated as a crucial parameter in the Isolation Forest algorithm to achieve the automated identification of bad channels. To evaluate the effectiveness of our method, we employed three accuracy metrics and conducted comparative assessments via the following methods: box-plot, density-based spatial clustering of applications with noise, and local outlier factor. Our approach is validated via simulation experiments and real measurements using OPM-magnetoencephalography data. The efficiency of subsequent OPM-MEG preprocessing and source location analysis is improved.

Honey-block: Edge assisted ensemble learning model for intrusion detection and prevention using defense mechanism in IoT

The Internet of Things (IoT) has gained popularity with interconnected devices and diverse network applications, leading to increased vulnerability of sensitive data to security threats. Many researchers have focused on intrusion detection without considering prevention mechanisms. To overcome these issues, we propose the honeypot and blockchain-based intrusion detection and prevention (HB-IDP) model, in which edge computing is introduced to reduce the latency during communication. Initially, three-fold authentication is performed for entities (users, devices, and gateway) to ensure legitimacy using the camellia encryption algorithm (CEA), which provides secret keys. The proposed datasets (i.e., UNSW-NB15 and BoT-IoT) are pre-processed at the gateway using min–max normalization to reduce redundancy and complexity during feature extraction and classification. Signature-based intrusion detection is performed on the pre-processed data, with known attacks classified into three classes (normal, malicious, and suspicious) using the improved isolation forest (IIF) algorithm. Suspicious data are forwarded for anomaly detection to the edge level; here, a honeypot is deployed to attract the attacker’s patterns. Ensemble learning technique, including multi-layer perceptron (MLP), general adversarial network (GAN), and lightweight convolutional neural Network (LCNN), is applied to classify suspicious packet behaviors. Once intrusions are detected, the proposed work prevents future intrusions by generating reports, which are then encrypted by the CEA algorithm and provided to legitimate users. All transactions (i.e., key generation, report generation, and attacker patterns) are stored in the blockchain. The HB-IDP model’s performance and effectiveness were evaluated using network simulator 3.26 (NS-3.26), showcasing its superiority over existing approaches.

Optimizing the Isolation Forest Algorithm for Identifying Abnormal Behaviors of Students in Education Management Big Data

With the changes in educational models, applying computer algorithms and artificial intelligence technologies to data analysis in universities has become a research hotspot in the field of intelligent education. In response to the increasing amount of student data in universities, this study proposes to use an optimized isolated forest algorithm for recognizing features to detect abnormal student behavior concealed in big data for educational management. Firstly, it uses logistic regression algorithm to update the calculation method of isolated forest weights, and then uses residual statistics to eliminate redundant forests. Finally, it utilizes discrete particle swarm optimization to optimize the isolated forest algorithm. On this basis, improvements have also been made to the traditional gated loop unit network. It merges the two improved algorithm models and builds an anomaly detection model for collecting college student education data. The experiment shows that the optimized isolated forest algorithm has a recognition accuracy of 0.986 and a training time of 1 second. The recognition accuracy of the improved gated loop unit network is 0.965, and the training time is 0.16 seconds. In summary, the constructed model can effectively identify abnormal data of college students, thereby helping educators to detect students' problems in time and helping students to improve their learning status.

A lightweight unsupervised adversarial detector based on autoencoder and isolation forest

Although deep neural networks (DNNs) have performed well on many perceptual tasks, they are vulnerable to adversarial examples that are generated by adding slight but maliciously crafted perturbations to benign images. Adversarial detection is an important technique for identifying adversarial examples before they are entered into target DNNs. Previous studies that were performed to detect adversarial examples either targeted specific attacks or required expensive computation. Designing a lightweight unsupervised detector is still a challenging problem. In this paper, we propose an AutoEncoder-based Adversarial Examples (AEAE) detector that can guard DNN models by detecting adversarial examples with low computation in an unsupervised manner. The AEAE includes only a shallow autoencoder that performs two roles. First, a well-trained autoencoder has learned the manifold of benign examples. This autoencoder can produce a large reconstruction error for adversarial images with large perturbations, so we can detect significantly perturbed adversarial examples based on the reconstruction error. Second, the autoencoder can filter out small noises and change the DNN’s prediction on adversarial examples with small perturbations. It helps to detect slightly perturbed adversarial examples based on the prediction distance. To cover these two cases, we utilize the reconstruction error and prediction distance from benign images to construct a two-tuple feature set and train an adversarial detector using the isolation forest algorithm. We show empirically that AEAE is an unsupervised and inexpensive detector against most state-of-the-art attacks. Through the detection in these two cases, there is nowhere to hide adversarial examples.

Automatic detection of crop lodging from multitemporal satellite data based on the isolation forest algorithm

Rapid identification of crop lodging has implications for optimizing disaster prevention and mitigation measures. This study proposes a sequentially integrated algorithm for crop lodging identification, which first automatically extracts lodged and non-lodged samples based on the isolation forest algorithm from the screened features dataset and then uses a supervised classifier for lodging identification. Taking the crop lodging disaster caused by the super typhoon “Maysak” in Bohetai Township as an example, we calculated 162 feature indices from Sentinel-1 and Sentinel-2 satellite data pre- and post-disaster and screened the optimal features. Then, the abnormal scores of the cultivated-land pixels were calculated using the isolation forest algorithm. After that, we calculated the coefficient of variation (CV) of the abnormal scores of each group, extracted the lodged and non-lodged samples based on the cumulative CV contribution rate, and input samples into three supervised classifiers (random forest, maximum likelihood, and support vector machine) to identify lodged crop pixels. The results showed that feature screening increased the identification accuracy of crop lodging by more than 10% and difference features of short-wave infrared (SWIR) and red edge bands contributed significantly to the identification. The accuracy of crop lodging automatic identification was verified using samples obtained from visual interpretation and digitization of UAV images as reference data and the overall accuracy reached 78%. The proposed method can simultaneously recognize lodging of different crop types without the need to select lodged samples and extract lodged ranges for each crop. With low data acquisition costs and high automation, the proposed method has promising applications in agricultural disaster emergency management and agricultural insurance loss assessment.

Missing data filling in soft sensing using denoising diffusion probability model

With the aim of addressing the problem of degradation in soft measurement accuracy due to missing data in industrial processes, a filling method based on the denoising diffusion probability model (DDPM) is proposed here to improve the accuracy of soft measurement modeling. First, missing regions are detected with the help of an improved Isolation Forest algorithm to obtain information such as the locations and numbers of missing data regions. Next, a data generation model is constructed based on DDPM and new samples are obtained. By adjusting the threshold for normal operation of the system and the weight sampler, filler samples that are similar to the distribution of the original data can be filtered from the new samples to form a complete dataset. The feasibility of the proposed missing data filling method is explored through numerical simulations, and its superiority in terms of improving the prediction accuracy of soft measurements is verified in regard to the nickel flash smelting process.

Digital Twin-Based Fault Diagnosis Platform for Final Rolling Temperature in Hot Strip Production.

The final rolling temperature in hot rolling is an important process parameter for hot-rolled strips and greatly influences their mechanical properties and rolling stability. The diagnosis of final rolling temperature anomalies in hot rolling has always been difficult in industry. A data-driven risk assessment method for detecting final rolling temperature anomalies is proposed. In view of the abnormal setting value for the strip head, a random forest model is established to screen the process parameters with high feature importance, and the isolation forest algorithm is used to evaluate the risk associated with the remaining parameters. In view of the abnormal process curve of the full length of the strip, the Hausdorff distance algorithm is used to eliminate samples with large deviations, and a risk assessment of the curve is carried out using the LCSS algorithm. Aiming to understand the complex coupling relationship between the influencing factors, a method for identifying the causes of anomalies, combining a knowledge graph and a Bayesian network, is established. According to the results of the strip head and the full-length risk assessment model, the occurrence of the corresponding nodes in the Bayesian network is determined, and the root cause of the abnormality is finally output. By combining mechanistic modeling and data modeling techniques, it becomes possible to rapidly, automatically, and accurately detect and analyze final rolling temperature anomalies during the rolling process. When applying the system in the field, when compared to manual analysis by onsite personnel, the accuracy of deducing the causes of anomalies was found to reach 92%.

A Hybrid Approach for Soil Total Nitrogen Anomaly Detection Integrating Machine Learning and Spatial Statistics

Soil total nitrogen is one of the most important basic indicators for fertiliser decision making, but tens of millions of soil total nitrogen sampling data have been accumulated, forming a huge database. In this large database, there is a large amount of anomalous data, which can interfere with data analysis, affect the construction of spatial interpolation and prediction models, and then affect the accuracy of nutrient management decisions. The traditional method of identifying soil total nitrogen anomalies based on boxplots suffers from the problems of not being able to identify local anomalies, which can easily lead to misclassification of soil total nitrogen data anomalies, and the detection efficiency is not high. We propose a method to identify soil total nitrogen outliers by combining the Isolation Forest algorithm and local spatial autocorrelation analysis, which can simultaneously detect global and local outliers from large amounts of data and combine organic matter as an auxiliary indicator in the spatial analysis to help judge local outliers. Finally, the results of global and local anomalies were combined to provide a comprehensive assessment of the soil nitrogen data, avoiding the misjudgement or omission of judgement that can occur when using a single method. Using 25,930 soil test data from Yunnan Province in 2009 as an example, we compared and analysed the typical boxplot method and the unsupervised OneClassSVM method and evaluated the performance of each method in terms of correct detection rate, false positive rate and false negative rate. The results show that the proposed method has a correct detection rate (TR) of 99.97%, a false positive rate (FPR) of 8.06% and a false negative rate (FNR) of 0.01% on the data, which shows high validity and accuracy; it is also comparable to the independent isolated forests (FNR = 4.76%), boxplot (FNR = 3.90%) and OneClassSVM (FNR = 4.77%), and the false negative rate is reduced by 4.75%, 3.89% and 4.76%, respectively.

Integrated Energy System Based on Isolation Forest and Dynamic Orbit Multivariate Load Forecasting

Short-term load forecasting is a prerequisite for achieving intra-day energy management and optimal scheduling in integrated energy systems. Its prediction accuracy directly affects the stability and economy of the system during operation. To improve the accuracy of short-term load forecasting, this paper proposes a multi-load forecasting method for integrated energy systems based on the Isolation Forest and dynamic orbit algorithm. First, a high-dimensional data matrix is constructed using the sliding window technique and the outliers in the high-dimensional data matrix are identified using Isolation Forest. Next, the hidden abnormal data within the time series are analyzed and repaired using the dynamic orbit algorithm. Then, the correlation analysis of the multivariate load and its weather data is carried out by the AR method and MIC method, and the high-dimensional feature matrix is constructed. Finally, the prediction values of the multi-load are generated based on the TCN-MMoL multi-task training network. Simulation analysis is conducted using the load data from a specific integrated energy system. The results demonstrate the proposed model’s ability to significantly improve load forecasting accuracy, thereby validating the correctness and effectiveness of this forecasting approach.

MLAnalysis: An open-source program for high energy physics analyses

We present a python-based program for phenomenological investigations in particle physics using machine learning algorithms, called MLAnalysis. The program is able to convert LHE and LHCO files generated by MadGraph5_aMC@NLO into data sets for machine learning algorithms, which can analyze the information of the events. At present, it contains three machine learning (ML) algorithms: isolation forest (IF) algorithm, nested isolation forest (NIF) algorithm, kmeans anomaly detection (KMAD), and some basic functionality to analyze the kinematic features of a data set. Users can use this program to improve the efficiency of searching for new physics signals. Program summaryProgram Title: MLAnalysisCPC Library link to program files:https://doi.org/10.17632/xnrgv2z76h.1Developer's repository link:https://github.com/NBAlexis/MLAnalysisCode Ocean capsule:https://codeocean.com/capsule/3663195Programming language: Python3.8 and aboveNature of problem: With the continuous accumulation of experimental data, the research of high energy physics needs to process a large amount of data. ML methods can help us to improve the effect and efficiency of data analysis. Converting the data from experiments or Monte Carlo (MC) simulated events into data sets available for ML has become an important requirement. A program platform is needed for data preparation, as well as the application of various ML algorithms to improve the selection capability of target events and the efficiency of particle identification.Solution method: Supply an event analysis platform that supports ML approaches. The program is able to convert LHE and LHCO files into data sets that can be used for ML algorithms, and apply data preparation. In the data preparation step, the program transforms the raw data into a format that can be used to train and test machine learning algorithms, optimizes the adaptabilities and generalization capabilities of algorithms. The program offers several algorithms, including IF, NIF, and KMAD, which provide NP model independent and standard model effective field theory operator independent methods to optimize event selection strategies.

Shear strength prediction of FRP-strengthened concrete beams using interpretable machine learning

This study identified the factors affecting the contribution of externally bonded fiber reinforced polymer (FRP) composite (Vf) to the shear strength of reinforced concrete (RC) beams with internal shear reinforcement through the use of interpretable machine learning (ML). A comprehensive database with 442 RC beams strengthened in shear with FRP was established and subjected to data anomaly detection using the isolation forest algorithm. Six ML models (artificial neural networks (ANN), XGBoost, random forest, CatBoost, LightGBM, AdaBoost algorithms) were trained to predict Vf. The ML models outperform six traditional explicit equations commonly used in design codes or the literature in terms of accuracy and variation. Based on the interpretable ML, the important variables were revealed; in particular the effective height of FRP, shear span ratio, and reinforcement method (U-wrap, full wrapping, side bonding) had significant influence on Vf. Finally, driven by results of the ML models, an explicit equation was derived to predict Vf, considering the physical meaning of the key influencing parameters. This study combines ML models and traditional physical models to achieve a novel, interpretable ML method to predict Vf.

Loitering behavior detection by spatiotemporal characteristics quantification based on the dynamic features of Automatic Identification System (AIS) messages.

The capability of the Automatic Identification System (AIS) to provide real-time worldwide coverage of ship tracks has made it possible for maritime authorities to utilize AIS as a means of surveillance to identify anomalies. Anomaly detection in maritime traffic is crucial as anomalous behavior may be a sign of either emergencies or illegal activities. Anomalous ships are recognized based on their behavior by manual examination. Such work requires extensive effort, especially for nationwide surveillance. To deal with this, researchers proposed computational methods to analyze vessel behavior. However, most approaches are region-dependent and require a profile of normality to detect anomalies, and amongst the six types of anomaly, loitering is the least explored. Loitering is not necessarily anomalous behavior as it is common for certain types of ships, such as pilot boats and research vessels. However, tankers and cargo ships normally do not engage in loitering. Based on 12-month manually examined data, nearly 60% of the identified anomalies were loitering, particularly for those of types cargo and tanker. Although manual identification is inefficient, automatically identifying abnormal vessels by merely implementing computing algorithms is not yet feasible. It still needs subject matter experts' assessments. This study proposes a region-independent method to automatically detect loitering without training normal instances and produces a ranked list of loitering vessels to facilitate further anomaly investigation. First, the loitering spatiotemporal characteristics are defined: (1) movement of frequent course change, with a certain speed, within a certain spatial range, (2) movement of frequent course change within traversed geodetic distance, (3) might demonstrate frequent extreme turning, and (4) extreme turning produces a significant discrepancy between the course over ground and the heading of the ship. Then, the characteristics are quantified by manipulating the dynamic information of AIS messages. Finally, the parameters to determine a loitering trajectory are formulated by comparing the rate of course change, speed, and the discrepancy between heading and course with the area of spatial range enclosing the trajectory and the geodetic distance between the start and end point. The loitering score of each trajectory is calculated with the parameters, and the Isolation Forest algorithm is employed to establish a threshold and rank. Then, geographic visualization is created for intuitive evaluation. An experiment was conducted on a real-world dataset covering a sea area of 610,116.37 km2. The results prove the efficacy of the proposed method. It remarkably outperforms the existing approach with 97% accuracy and 92% F-score. The experiment produces a ranked list of loitering vessels and an intuitive visualization in the relevant geographic area. In the realworld scenario, they are practical means to support further examination by human operators.