Improvement In F1 Score Research Articles

Evaluating Synthetic Data Augmentation to Correct for Data Imbalance in Realistic Clinical Prediction Settings.

Predictive modeling holds a large potential in clinical decision-making, yet its effectiveness can be hindered by inherent data imbalances in clinical datasets. This study investigates the utility of synthetic data for improving the performance of predictive modeling on realistic small imbalanced clinical datasets. We compared various synthetic data generation methods including Generative Adversarial Networks, Normalizing Flows, and Variational Autoencoders to the standard baselines for correcting for class underrepresentation on four clinical datasets. Although results show improvement in F1 scores in some cases, even over multiple repetitions, we do not obtain statistically significant evidence that synthetic data generation outperforms standard baselines for correcting for class imbalance. This study challenges common beliefs about the efficacy of synthetic data for data augmentation and highlights the importance of evaluating new complex methods against simple baselines.

TransURL: Improving malicious URL detection with multi-layer Transformer encoding and multi-scale pyramid features

While machine learning progress is advancing the detection of malicious URLs, advanced Transformers applied to URLs face difficulties in extracting local information, character-level information, and structural relationships. To address these challenges, we propose a novel approach for malicious URL detection, named TransURL, that is implemented by co-training the character-aware Transformer with three feature modules—Multi-Layer Encoding, Multi-Scale Feature Learning, and Spatial Pyramid Attention. This special Transformer allows TransURL to extract embeddings that contain character-level information from URL token sequences, with three feature modules contributing to the fusion of multi-layer Transformer encodings and the capture of multi-scale local details and structural relationships. The proposed method is evaluated across several challenging scenarios, including class imbalance learning, multi-classification, cross-dataset testing, and adversarial sample attacks. The experimental results demonstrate a significant improvement compared to the best previous methods. For instance, it achieved a peak F1-score improvement of 40% in class-imbalanced scenarios, and exceeded the best baseline result by 14.13% in accuracy in adversarial attack scenarios. Additionally, we conduct a case study where our method accurately identifies all 30 active malicious web pages, whereas two pior SOTA methods miss 4 and 7 malicious web pages respectively. The codes and data are available at: https://github.com/Vul-det/TransURL/.

Integration of Relation Filtering and Multi-Task Learning in GlobalPointer for Entity and Relation Extraction

The rise of knowledge graphs has been instrumental in advancing artificial intelligence (AI) research. Extracting entity and relation triples from unstructured text is crucial for the construction of knowledge graphs. However, Chinese text has a complex grammatical structure, which may lead to the problem of overlapping entities. Previous pipeline models have struggled to address such overlap problems effectively, while joint models require entity annotations for each predefined relation in the set, which results in redundant relations. In addition, the traditional models often lead to task imbalance by overlooking the differences between tasks. To tackle these challenges, this research proposes a global pointer network based on relation prediction and loss function improvement (GPRL) for joint extraction of entities and relations. Experimental evaluations on the publicly available Chinese datasets DuIE2.0 and CMeIE demonstrate that the GPRL model achieves a 1.2–26.1% improvement in F1 score compared with baseline models. Further, experiments of overlapping classification conducted on CMeIE have also verified the effectiveness of overlapping triad extraction and ablation experiments. The model is helpful in identifying entities and relations accurately and can reduce redundancy by leveraging relation filtering and the global pointer network. In addition, the incorporation of a multi-task learning framework balances the loss functions of multiple tasks and enhances task interactions.

Federated Learning Using Multi-Modal Sensors with Heterogeneous Privacy Sensitivity Levels

Data from multi-modal sensors, such as RGB cameras, thermal cameras, microphones, and mmWave radars, have gradually been adopted in various classification problems for better accuracy. Some sensors, like RGB cameras and microphones, however, capture privacy-invasive data, which are less likely to be used in centralized learning. Although the Federated Learning (FL) paradigm frees clients from sharing their sensor data, doing so results in reduced classification accuracy and increased training time. In this article, we introduce a novel Heterogeneous Privacy Federated Learning (HPFL) paradigm to better capitalize on the less privacy-invasive sensor data, such as thermal images and mmWave point clouds, by uploading them to the server for closing the performance gap between FL and centralized learning. HPFL not only allows clients to keep the more privacy-invasive sensor data private, such as RGB images and human voices, but also gives each client total freedom to define the levels of their privacy concern on individual sensor modalities. For example, more sensitive users may prefer to keep their thermal images private, while others do not mind sharing these images. We carry out extensive experiments to evaluate the HPFL paradigm using two representative classification problems: semantic segmentation and emotion recognition. Several key findings demonstrate the merits of HPFL: (i) compared to FedAvg, it improves foreground accuracy by 18.20% in semantic segmentation and boosts the F1-score by 4.20% in emotion recognition, (ii) with heterogeneous privacy concern levels, it achieves an even larger F1-score improvement of 6.17–16.05% in emotion recognition, and (iii) it also outperforms the state-of-the-art FL approaches by 12.04–17.70% in foreground accuracy and 2.54–4.10% in F1-score.

A Geometric Approach to Textual Augmented Data Filtering

Abstract Data augmentation is necessary if the amount of training data is insufficient for supervised learning. For natural language processing tasks, obtaining good quality augmented data is not easy. This paper introduces GATFilter, a novel method for filtering out inappropriate augmented textual data for text classification (TC). Utilizing geometric concepts, more specifically the principle component and convex hull analyses, this method adeptly preserves the semantic integrity of words within augmented texts. GATFilter is versatile and applicable across various types of textual augmentation methods. Experiments using several datasets and augmentation strategies showed that classifiers trained with GATFilter-filtered augmented data sets showed improvements in key performance metrics, including accuracy, precision, recall, and F1 score. The method’s efficacy is notably influenced by the quality of the underlying augmentation techniques, indicating its potential to complement and refine various text augmentation strategies. Furthermore, our analysis showed that GATFilter is particularly able to amplify the effectiveness of methods that generate good quality augmented data. GATFilter is openly available online on Github1, and as a Python package2.

Automated segmentation and deep learning classification of ductopenic parotid salivary glands in sialo cone-beam CT images.

This study addressed the challenge of detecting and classifying the severity of ductopenia in parotid glands, a structural abnormality characterized by a reduced number of salivary ducts, previously shown to be associated with salivary gland impairment. The aim of the study was to develop an automatic algorithm designed to improve diagnostic accuracy and efficiency in analyzing ductopenic parotid glands using sialo cone-beam CT (sialo-CBCT) images. We developed an end-to-end automatic pipeline consisting of three main steps: (1) region of interest (ROI) computation, (2) parotid gland segmentation using the Frangi filter, and (3) ductopenia case classification with a residual neural network (RNN) augmented by multidirectional maximum intensity projection (MIP) images. To explore the impact of the first two steps, the RNN was trained on three datasets: (1) original MIP images, (2) MIP images with predefined ROIs, and (3) MIP images after segmentation. Evaluation was conducted on 126 parotid sialo-CBCT scans of normal, moderate, and severe ductopenic cases, yielding a high performance of 100% for the ROI computation and 89% for the gland segmentation. Improvements in accuracy and F1 score were noted among the original MIP images (accuracy: 0.73, F1 score: 0.53), ROI-predefined images (accuracy: 0.78, F1 score: 0.56), and segmented images (accuracy: 0.95, F1 score: 0.90). Notably, ductopenic detection sensitivity was 0.99 in the segmented dataset, highlighting the capabilities of the algorithm in detecting ductopenic cases. Our method, which combines classical image processing and deep learning techniques, offers a promising solution for automatic detection of parotid glands ductopenia in sialo-CBCT scans. This may be used for further research aimed at understanding the role of presence and severity of ductopenia in salivary gland dysfunction.

Self‐training with Bayesian neural networks and spatial priors for unsupervised domain adaptation in crack segmentation

AbstractThis study proposes a novel self‐training framework for unsupervised domain adaptation in the segmentation of concrete wall cracks using accumulated crack data. The proposed method incorporates Bayesian neural networks for uncertainty estimation of pseudo‐labels, and spatial priors of cracks for screening noisy labels. Experiments demonstrate that the proposed approach achieves significant improvements in F1 score. Comparing the F1 scores, Bayesian DeepLabv3+ and Bayesian U‐Net showed performance improvements of 0.0588 and 0.1501, respectively, after domain adaptation. Furthermore, the integration of Stable Diffusion for few‐shot image generation enhances domain adaptation performance by 0.0332. The proposed framework enables high‐precision crack segmentation with as few as 100 target images, which can be easily obtained at the site, reducing the cost of model deployment in infrastructure maintenance. The study also investigates the optimal number of iterations for domain adaptation based on the uncertainty score, providing insights for practical implementation. The proposed method contributes to the development of efficient and automated structural health monitoring using AI.

Deep learning-based electricity theft prediction in non-smart grid environments

In developing countries, smart grids are nonexistent, and electricity theft significantly hampers power supply. This research introduces a lightweight deep-learning model using monthly customer readings as input data. By employing careful direct and indirect feature engineering techniques, including Principal Component Analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), UMAP (Uniform Manifold Approximation and Projection), and resampling methods such as Random-Under-Sampler (RUS), Synthetic Minority Over-sampling Technique (SMOTE), and Random-Over-Sampler (ROS), an effective solution is proposed. Previous studies indicate that models achieve high precision, recall, and F1 score for the non-theft (0) class, but perform poorly, even achieving 0 %, for the theft (1) class. Through parameter tuning and employing Random-Over-Sampler (ROS), significant improvements in accuracy, precision (89 %), recall (94 %), and F1 score (91 %) for the theft (1) class are achieved. The results demonstrate that the proposed model outperforms existing methods, showcasing its efficacy in detecting electricity theft in non-smart grid environments.

A hybrid network TEdgeNeXt for data-limited and resource-constrained fault diagnosis

In the field of intelligent machinery fault diagnosis, overcoming challenges arising from scarce labeled data and the demand for deployment on resource-constrained edge devices is imperative. To address these hurdles, this work aims to improve the ability of deep learning models to learn strong feature representations from limited data, while also reducing the model complexity. We presenting a novel network named TEdgeNeXt, the approach begins with a new signal-to-image conversion method, which is proved to be able to acquire less training data quantity. Structurally, the Convolutional (Conv.) Encoder initially is employed with depth-wise separable convolution to control the size of model rather than the traditional convolution, and the Split Depth-wise Transpose Attention (SDTA) encoder is consequently utilized by leveraging a multidimensional processing approach and the Multi-head Self-Attention which is across the channel dimensions instead of the spatial channel. By doing so, it effectively handles challenges such as high multiply-additions (MAdds) and increased latency through Flops and params. On the other hand, the fine-tune-based transfer learning technique is able to be extended in our approach for improving the capacity of generalizing. Ultimately, it indicates the noticeable improvements in Top-1 Accuracy (T1A), Mean Precision (MP), Mean Recall (MR), and Mean F1 score (MF1) across three distinct datasets.

The robust SmartEnsembleNet: A game changer in finger knuckle biometrics

SmartEnsembleNet represents a major leap forward in the domain of biometric identification, focusing on the relatively underexplored area of finger knuckle biometrics. This research enhances the effectiveness and uniqueness of biometric systems by leveraging the distinct and secure features of finger knuckle patterns. As a complement to existing biometric techniques like fingerprints and facial recognition, it adds an additional layer of security and reliability. Addressing the challenges faced by finger knuckle identification, such as image acquisition, environmental effects, and aging, SmartEnsembleNet introduces a pioneering multi-layered ensemble learning strategy that combines eight diverse machine learning models. These models, trained via a 5-fold cross-validation on datasets from IIT Delhi and PolyU, demonstrate significant improvements in accuracy, precision, recall, and F1-score. A distinctive feature of SmartEnsembleNet is its innovative meta-model, streamlining the creation of a meta-dataset. This dataset is created by applying bilinear interpolation to the predicted probabilities from the top-performing model, and it is further refined by subtracting these interpolated probabilities from the validation data. This iterative process generates ‘smart data’, which serves as the basis for training the ultimate ‘smart model’. Logistic regression is chosen for its outstanding performance. This method, applied uniformly to both training and testing phases, considerably enhances classification accuracy in finger knuckle biometrics, establishing new benchmarks. SmartEnsembleNet’s performance undergoes a detailed evaluation using the Friedman test and Nemenyi post-hoc analysis, guaranteeing a selection process grounded in statistical rigor. With its resilience to noise and adaptability under various conditions, SmartEnsembleNet not only addresses existing gaps in finger knuckle biometric research but also pioneers future advancements, providing a versatile and secure solution impactful for both academia and high-security scenarios.

SCC-GPT: Source Code Classification Based on Generative Pre-Trained Transformers

Developers often rely on online resources, such as Stack Overflow (SO), to seek assistance for programming tasks. To facilitate effective search and resource discovery, manual tagging of questions and posts with the appropriate programming language is essential. However, accurate tagging is not consistently achieved, leading to the need for the automated classification of code snippets into the correct programming language as a tag. In this study, we introduce a novel approach to automated classification of code snippets from Stack Overflow (SO) posts into programming languages using generative pre-trained transformers (GPT). Our method, which does not require additional training on labeled data or dependency on pre-existing labels, classifies 224,107 code snippets into 19 programming languages. We employ the text-davinci-003 model of ChatGPT-3.5 and postprocess its responses to accurately identify the programming language. Our empirical evaluation demonstrates that our GPT-based model (SCC-GPT) significantly outperforms existing methods, achieving a median F1-score improvement that ranges from +6% to +31%. These findings underscore the effectiveness of SCC-GPT in enhancing code snippet classification, offering a cost-effective and efficient solution for developers who rely on SO for programming assistance.

Low-resource entity resolution with domain generalization and active learning

Entity Resolution (ER), a fundamental task in data cleaning and integration, is critical in various fields such as healthcare, e-commerce, and social networks. Traditional ER methods are constrained by the need for substantial labeled samples and the challenge of generalization to unseen domains. To address the low-resource challenge in ER, a novel two-phase framework is proposed. Initially, we introduce the Domain Generalization Entity Resolution (DGER) framework, combining domain adversarial learning and simulated target learning to improve the generalization performance on unseen target domains. Subsequently, to further adapt to the target dataset, we present a novel active learning approach called Domain-Aware Uncertainty Active Learning (DUAL), for fine-tuning the DGER model with minimal annotation cost. DUAL manually annotates target domain samples that are highly uncertain and exhibit high divergence from the source, while assigning pseudo-labels to high-confidence samples. Experimental results on multiple real-world datasets demonstrate that our framework outperforms traditional ER methods in generalizing to unseen domains. Specifically, our DGER method outperforms the best-performing ER baseline in each task, achieving an average F1 score improvement of 9.02% across eight different test tasks. Moreover, within a limited annotation budget during the active learning phase, our DUAL fine-tuning strategy for the ER model outperforms uncertainty-based active learning techniques.

A combined EEG motor and speech imagery paradigm with automated successive halving for customizable command selection

ABSTRACT The classification performance of endogenous electroencephalogram (EEG) brain-computer interfaces (BCIs) can be improved by hybridizing the paradigm through the use of commands from multiple paradigms. Hybrid paradigms using motor imagery (MI) and speech imagery (SI) have shown promise, but there is a lack of research into: i) their effectiveness when compared to pure MI and SI for multiclass problems, and ii) automated command selection. This study investigates multiclass MI and SI hybrid paradigms and compares the results to those obtained using pure MI and SI. Performance was assessed using F1 score and accuracy. The performances of all possible hybrid paradigm designs were assessed. The analysis indicated that hybridization does not always guarantee improved performance when compared to the pure paradigms, and there is inter-subject variation in the best paradigm. This confirmed the need for automated subject-specific hybrid paradigm designs. An automated hybrid paradigm selection technique using successive halving (SH) for expedited computational times was developed and results were compared to those obtained using a standard grid search. The SH approach resulted in an improvement in F1 score of 21.09% and 36.86% compared to MI and SI and led to a reduction in computational times of 82.80% compared to grid search.

Multi-Object Detection and Tracking with Modified Optimization Classification in Video Sequences

The paper presents a novel approach to enhancing multi-object detection and tracking in video sequences using a Modified Ant Swarm Optimization Deep Learning (ASO-DL) algorithm. The ASO-DL algorithm synergistically combines the optimization capabilities of ant swarm optimization with the powerful feature extraction abilities of deep learning models, resulting in a robust framework for realtime video analytics. Extensive simulations and experiments demonstrate significant improvements in key performance metrics, including accuracy, precision, recall, and F1 score, across various iterations. The proposed method consistently outperforms baseline models, achieving a final best fitness value of 0.96, with an accuracy of 0.98, precision of 0.99, and recall of 0.95. Additionally, classification results across different datasets such as CIFAR-10, IMDB, COCO, and ImageNet highlight the algorithm’s versatility and effectiveness. This research contributes to the field by providing a highly optimized solution for complex multi-object tracking tasks, offering substantial advancements in the accuracy and efficiency of real-time object detection systems. The findings hold significant potential for applications in surveillance, autonomous vehicles, and other domains requiring precise and reliable multi-object tracking.

Resolving intra-repeat variation in medically relevant VNTRs from short-read sequencing data using the cardiovascular risk gene LPA as a model

BackgroundVariable number tandem repeats (VNTRs) are highly polymorphic DNA regions harboring many potentially disease-causing variants. However, VNTRs often appear unresolved (“dark”) in variation databases due to their repetitive nature. One particularly complex and medically relevant VNTR is the KIV-2 VNTR located in the cardiovascular disease gene LPA which encompasses up to 70% of the coding sequence.ResultsUsing the highly complex LPA gene as a model, we develop a computational approach to resolve intra-repeat variation in VNTRs from largely available short-read sequencing data. We apply the approach to six protein-coding VNTRs in 2504 samples from the 1000 Genomes Project and developed an optimized method for the LPA KIV-2 VNTR that discriminates the confounding KIV-2 subtypes upfront. This results in an F1-score improvement of up to 2.1-fold compared to previously published strategies. Finally, we analyze the LPA VNTR in > 199,000 UK Biobank samples, detecting > 700 KIV-2 mutations. This approach successfully reveals new strong Lp(a)-lowering effects for KIV-2 variants, with protective effect against coronary artery disease, and also validated previous findings based on tagging SNPs.ConclusionsOur approach paves the way for reliable variant detection in VNTRs at scale and we show that it is transferable to other dark regions, which will help unlock medical information hidden in VNTRs.

High-precision flood detection and mapping via multi-temporal SAR change analysis with semantic token-based transformer

Flood detection in crisis and disaster management is significantly facilitated by the analysis of synthetic aperture radar (SAR) imagery. Traditional flood detection techniques focus more on SAR image pairs than on the optical level. However, the distinctive characteristics of SAR images, characterized by limited visual information, pervasive speckle noise, and analogous backscatter signals, present formidable obstacles to accurately identifying water bodies and extracting change features. Consequently, the performance of existing methods remains unsatisfactory. This paper addresses this challenge by focusing on disparities between SAR image pairs and introducing a pioneering semantic token-based transformer network, denoted as SemT-Former, to enhance flood detection accuracy. SemT-Former operates by prioritizing changes of interest rather than fully comprehending the entire image scene. This is achieved through the integration of temporal-wise feature representation and the introduction of a class token to capture high-level segmentation associated with changes in water bodies. These innovations augment the model’s capacity to discriminate between genuine changes in water bodies and spurious changes induced by similar signals or speckle noise. The effectiveness of SemT-Former is evaluated through a case study in Khartoum, Sudan, focusing on flood detection and the estimation of damaged farmland near river confluences. Experimental results demonstrate that SemT-Former outperforms existing methods, exhibiting a 90.6% improvement in F1-score and an 88.5% enhancement in IoU. This underscores SemT-Former as a promising solution for precise and effective flood mapping from SAR images.

Encoding temporal information in deep convolution neural network.

A recent development in deep learning techniques has attracted attention to the decoding and classification of electroencephalogram (EEG) signals. Despite several efforts to utilize different features in EEG signals, a significant research challenge is using time-dependent features in combination with local and global features. Several attempts have been made to remodel the deep learning convolution neural networks (CNNs) to capture time-dependency information. These features are usually either handcrafted features, such as power ratios, or splitting data into smaller-sized windows related to specific properties, such as a peak at 300 ms. However, these approaches partially solve the problem but simultaneously hinder CNNs' capability to learn from unknown information that might be present in the data. Other approaches, like recurrent neural networks, are very suitable for learning time-dependent information from EEG signals in the presence of unrelated sequential data. To solve this, we have proposed an encoding kernel (EnK), a novel time-encoding approach, which uniquely introduces time decomposition information during the vertical convolution operation in CNNs. The encoded information lets CNNs learn time-dependent features in addition to local and global features. We performed extensive experiments on several EEG data sets-physical human-robot collaborations, P300 visual-evoked potentials, motor imagery, movement-related cortical potentials, and the Dataset for Emotion Analysis Using Physiological Signals. The EnK outperforms the state of the art with an up to 6.5% reduction in mean squared error (MSE) and a 9.5% improvement in F1-scores compared to the average for all data sets together compared to base models. These results support our approach and show a high potential to improve the performance of physiological and non-physiological data. Moreover, the EnK can be applied to virtually any deep learning architecture with minimal effort.

Label-semantics enhanced multi-layer heterogeneous graph convolutional network for Aspect Sentiment Quadruplet Extraction

Aspect-based sentiment analysis (ABSA) is a pivotal area within natural language processing (NLP) that focuses on extracting fine-grained sentiment information from text. A particularly demanding task within ABSA is Aspect Sentiment Quadruplet Extraction (ASQE), which aims to identify quadruplets comprising aspect terms, their corresponding opinion terms, sentiment polarity, and categories. This level of analysis is crucial for downstream applications such as sentiment monitoring and user experience research, offering nuanced insights into textual data. However, current methodologies fall short of fully leveraging the rich linguistic features of sentences and the semantic information of labels embedded within sentences. To address this gap, this paper introduces a novel framework, the Label-Semantics Enhanced Multi-layer Heterogeneous Graph Convolutional Network (LSEMH-GCN), specifically designed for ASQE. Our model integrates sentence linguistic features and label semantics to construct a graph neural network tailored for this task. It employs a multi-layer graph convolutional network that synergizes various linguistic features, and utilizes Biaffine attention to enrich the label probability distribution for token pairs with semantic label information. Furthermore, our approach introduces a token pair vector concatenation strategy alongside an advanced asymmetric label tagging scheme to enhance quadruplet extraction. Comprehensive evaluations on benchmark datasets reveal that LSEMH-GCN significantly surpasses existing state-of-the-art models, establishing a new benchmark for ASQE. Our model achieves an average F1 score improvement of 15.52% and 12.30% on the Restaurant-ACOS and Laptop-ACOS datasets, respectively.

Predicting ICU Interventions: A Transparent Decision Support Model Based on Multivariate Time Series Graph Convolutional Neural Network.

In this study, we present a novel approach for predicting interventions for patients in the intensive care unit using a multivariate time series graph convolutional neural network. Our method addresses two critical challenges: the need for timely and accurate decisions based on changing physiological signals, drug administration information, and static characteristics; and the need for interpretability in the decision-making process. Drawing on real-world ICU records from the MIMIC-III dataset, we demonstrate that our approach significantly improves upon existing machine learning and deep learning methods for predicting two targeted interventions, mechanical ventilation and vasopressors. Our model achieved an accuracy improvement from 81.6% to 91.9% and a F1 score improvement from 0.524 to 0.606 for predicting mechanical ventilation interventions. For predicting vasopressor interventions, our model achieved an accuracy improvement from 76.3% to 82.7% and a F1 score improvement from 0.509 to 0.619. We also assessed the interpretability by performing an adjacency matrix importance analysis, which revealed that our model uses clinically meaningful and appropriate features for prediction. This critical aspect can help clinicians gain insights into the underlying mechanisms of interventions, allowing them to make more informed and precise clinical decisions. Overall, our study represents a significant step forward in the development of decision support systems for ICU patient care, providing a powerful tool for improving clinical outcomes and enhancing patient safety.

A criteria-based classification model using augmentation and contrastive learning for analyzing imbalanced statement data

Criteria Based Content Analysis (CBCA) is a forensic tool that analyzes victim statements. It involves the categorization of victims' statements into 19 distinct criteria classifications, playing a crucial role in evaluating the authenticity of testimonies by discerning whether they are rooted in genuine experiences or fabricated accounts. The exclusion of subjective opinions becomes imperative to assess statements through this forensic tool objectively. This study proposes developing an objective classification model for CBCA-based statement analysis using natural language processing techniques. Nevertheless, achieving optimal classification performance proves challenging due to imbalances in data distribution among the various criterion classifications. To enhance the accuracy and reliability of the classification model, this research employs data augmentation techniques and dual contrastive learning methods for fine-tuning the RoBERTa language model. Furthermore, model-based optimization techniques are also applied to identify augmented hyper-parameters and maximize the model's classification performance. The study's findings, including an 8.5% improvement in macro F1 score compared to human classification results, a 24% improvement in macro F1 score, and a 13% improvement in accuracy compared to previous human classification results, suggest that the proposed model is highly effective in reducing the influence of human subjectivity in statement analysis. The proposed model has significant implications for legal proceedings and criminal investigations, as it can provide a more objective and reliable method for evaluating the credibility of victim statements. Reducing human subjectivity in the statement analysis process can increase the accuracy of verdicts and help ensure that justice is served.