Deep Learning Techniques Research Articles

Research of Automatic Modulation Recognition Technology based on Deep Learning

Automatic modulation recognition is a vital component in communication systems. It helps in recognizing and classifying signal modulation automatically without the acquirement of the signals modulation type in advance. This paper aims to provide a comprehensive exploration of the application of deep learning techniques in the field of Automatic Modulation Recognition (AMR), with a specific emphasis on the analysis and comparison of two prominent neural network architectures: Convolutional Neural Networks (CNN) and Convolutional Long Short-Term Memory Networks (CLDNN). By delving into the structural and functional aspects of these models, the paper seeks to elucidate how CNNs, with their ability to capture spatial features through convolutional layers, and CLDNNs, which enhance this capability by integrating Long Short-Term Memory (LSTM) units to handle temporal dynamics, contribute to the advancement of AMR technology. We used the DeepSig dataset: RadioML 2018.01A to evaluate them, comparing their performance in different signal-to-noise ratio scenarios. The results manifested that CLDNN performed better than CNN obviously, especially in the low SNR scenario where CLDNN reached a higher validation set accuracy.

AI-Powered Waste Detection and Localization Using Region Proposal Network and CNN

To reduce environmental pollution caused by improper waste disposal, efficient waste management solutions are essential. Traditional methods rely on manual monitoring, which is time-consuming and ineffective. Advancements in artificial intelligence and deep learning enable automated waste detection and localization, improving efficiency in waste management. Faster R-CNN, an advanced object detection model, enhances precision in recognizing and localizing waste in visual data. By utilizing deep learning techniques, waste detection achieves high accuracy, reducing dependence on manual inspections.The integration of Faster R-CNN with image processing techniques allows for real-time identification of waste objects in complex environments. Localization is achieved by drawing bounding boxes around detected waste, providing precise positional information. The Region Proposal Network (RPN) plays a key role in this process by generating candidate object regions through anchor boxes and refining them based on object scores. Tensor Flow’s Object Detection API facilitates model training and optimization, ensuring accurate recognition of waste materials. High-performance convolutional neural networks enhance feature extraction, distinguishing waste from non-waste objects effectively. The use of deep learning in image-based waste detection supports scalable waste management solutions.Automated waste detection and localization contribute to sustainable urban development by enabling proactive waste management strategies. The ability to accurately identify and mark waste locations enhances the efficiency of cleaning processes and reduces environmental hazards. By analyzing waste distribution patterns, authorities can optimize resource allocation and improve waste disposal planning. The adoption of AI-driven approaches in waste detection minimizes human effort while promoting a cleaner and healthier environment. Key Words: AI-Driven waste detection, Faster R-CNN, Region Proposal Network, TensorFlow object detection API, Image processing and Localization.

Abstract 4186: Discovery of novel PARP1-selective inhibitors for treatment of brain tumors using artificial intelligence

Abstract Introduction: Brain tumors, including primary brain tumors and central nervous system (CNS) metastases, remain among the most challenging malignancies to treat, with therapeutic options often limited by the inability of drugs to penetrate the blood-brain barrier (BBB). Poly(ADP-ribose) polymerase (PARP) is a key enzyme in DNA repair, and inhibition of PARP1 specifically drives synthetic lethality in BRCA-mutated disease. While they have achieved commercial success, first generation PARP inhibitors are limited in their utility as they cannot readily pass through the blood-brain barrier and produce adverse side effects, likely driven by the collateral inhibition of PARP2. Development of a PARP1-selective CNS penetrant inhibitor could reduce toxicity, while providing a new therapeutic option for brain tumors. Traditional drug discovery methods are time-consuming and costly, necessitating innovative approaches. Here, we describe the application of Deep Docking combined with generative artificial intelligence (AI), to discover a novel, PARP1-selective inhibitor for use against brain tumors. Methods: Deep Docking utilizes deep learning to accelerate the prediction of the binding affinity of a large number of compounds to target proteins, streamlining the virtual screening process and allowing for rapid docking of billions of compounds against the PARP1 protein. Additionally, state-of-the-art generative algorithms and deep learning techniques for predicting CNS penetrance can be combined with Deep Docking to rapidly perform hit-to-lead optimization of a CNS-penetrating PARP1-selective inhibitor. Results: We will present Deep Docking screening results from billions of compounds. Validating in vitro and in vivo data, including PARP1 inhibition and selectivity, BBB permeability, metabolic stability, pharmacokinetic profile, and CNS penetration will be described. Conclusion: The Deep Docking platform is being used in concert with generative AI approaches to discover and finetune a selective PARP-1 inhibitor for brain tumors. This approach will not only improve the efficiency of drug discovery but also enhance the specificity and efficacy of potential therapeutics. Citation Format: Sarah Truong, Beibei Zhai, Louise Ramos, Mona Marzban, Fariba Ghaidi, Fuqiang Ban, Jason Smith, Mohit Pandey, Ekaterina Manskaia, Kally Singh, Hans Adomat, Xiaoqi Chen, John Langlands, Dennis Brown, Jeffrey Bacha, Colin Collins, Artem Cherkasov, Mads Daugaard. Discovery of novel PARP1-selective inhibitors for treatment of brain tumors using artificial intelligence [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4186.

Real-time Detection of Anomalous Trading Patterns in Financial Markets Using Generative Adversarial Networks

This paper presents a new methodology for real-time detection of malicious business models in financial markets using Generative Adversarial Networks (GANs). The proposed system combines deep learning techniques with special physical systems for real-time analysis of complex business processes while maintaining business-critical low-latency requirements. print many times. The framework uses a number of design elements that process business data flows at multiple time horizons, including business microstructure features and order book quality. A comprehensive analysis of data across 24 months of business data from multiple markets shows the foundation's performance, achieving 94.7% detection accuracy with sub-3ms latency. The system handles up to 150,000 transactions per second while maintaining stable performance across transactions. The framework's adaptive thresholding mechanism and hierarchical feature fusion approach reduce the vulnerability during high traffic times. Comparative analysis shows a 15.5% improvement in identification accuracy over traditional methods. This implementation includes a robust data pre-processing pipeline and an efficient computational model, enabling deployment to the manufacturing environment. These studies contribute to the advancement of business intelligence technology by introducing new applications of GANs in detecting real-world anomalies while solving critical problems. in processing business documents frequently.

Predicting Stock Prices with Investor Sentiment and Deep Learning Techniques

A key component of risk assessment, investment strategy, and financial planning is stock price prediction. Although technical indications and historical price patterns are the mainstays of traditional forecasting models, these frequently ignore the emotional and behavioural factors that affect financial markets. This study presents a hybrid methodology that combines investor sentiment gleaned from internet sources including social media and financial news sites with historical stock data. The approach improves the accuracy and responsiveness of predictions by incorporating both numerical trends and psychological cues. The system interprets textual data and produces sentiment scores by combining natural language processing methods with a deep learning architecture. Sentiment-enriched forecasts perform better than traditional models in both short-term trend prediction and market volatility awareness, according to experimental evaluation using real-world data. Practical ramifications for real-time forecasting systems and ethical issues are also covered in this paper. Keywords: Stock price, Investor Sentiment, Price Trend Prediction, LSTM, Financial Time Series, Machine learning

High-Risk Test Scenario Generation for Autonomous Vehicles at Roundabouts Using Naturalistic Driving Data

While autonomous vehicles have the potential to mitigate risks associated with dangerous driving behaviors, the safety and stability of autonomous driving technology in real-world applications still require comprehensive tests. Scenario-based virtual simulation testing has emerged as a crucial approach for testing autonomous vehicles, especially in critical scenarios under complex environments, such as merging scenarios at roundabouts with unique traffic features. However, the lack of high-risk scenarios in the real-world traffic domain presents challenges for simulation tests. To address these challenges, this study proposes a scenario generation framework for complex roundabout environments, focusing on merging areas, which is driven by trajectory data and employs generative deep learning techniques to create adversarial hazardous scenarios. Specifically, leveraging real trajectory data from roundabouts, the framework utilizes a time series generative adversarial network (TimeGAN) to generate realistic safety-critical driving trajectories. By creating specific hazardous scenarios, this strategy broadens the library of test scenarios and speeds up the testing process for autonomous vehicles. The significance of the scenarios produced is proven using Simulation of Urban Mobility (SUMO) and CARLA simulation, confirming their necessity in autonomous driving testing. The TimeGAN model effectively captures the spatial–temporal features of merging scenarios, generating high-quality data that enhance the testing scenario library. Findings of this study contribute to solving the problem of scarcity of critical scenarios in virtual testing and accelerate the testing procedure for self-driving automobiles.

Anomaly Detection in Financial Transactions Using Convolutional Neural Networks

The rise of digital financial systems has brought unprecedented convenience but has also exposed users and institutions to various fraudulent activities. Anomaly detection plays a critical role in ensuring financial security by identifying unusual transaction patterns that may indicate fraud or other irregularities. Traditional statistical and rule-based approaches, though effective to some extent, often fall short when dealing with the increasing volume and complexity of financial data. This study proposes a novel approach to anomaly detection in financial transactions using Convolutional Neural Networks (CNNs), a class of deep learning models primarily known for their success in image processing tasks. In this work, transactional data are preprocessed and transformed into structured formats suitable for CNN input. By treating sequences of financial transactions as temporal-spatial matrices, the CNN model learns intricate patterns that distinguish normal from anomalous behavior. Our methodology includes a comprehensive pipeline involving data normalization, feature engineering, and the construction of multi-channel representations to exploit CNNs' strength in hierarchical feature learning. We evaluate our model on benchmark financial datasets and compare its performance against traditional machine learning algorithms such as Support Vector Machines (SVM), Random Forest, and Logistic Regression. The CNN-based model demonstrates superior performance in terms of accuracy, precision, recall, and F1-score. Additionally, it shows robustness in detecting rare anomalies while minimizing false positives, a critical requirement in real-time financial fraud detection systems. The results indicate that CNNs can effectively capture both local and global dependencies within financial transaction sequences, making them suitable for large-scale and high-dimensional data environments. This study contributes to the growing body of research advocating for the adoption of deep learning techniques in financial anomaly detection and opens up possibilities for integrating CNNs with real-time monitoring systems for enhanced financial security. Future research may focus on hybrid models combining CNNs with recurrent layers to capture long-term dependencies more effectively.

RACF: A Multimodal Deep Learning Framework for Parkinson’s Disease Diagnosis Using SNP and MRI Data

The clinical diagnosis of Parkinson’s disease (PD) primarily relies on clinician-administered observational assessment tools, such as the Unified Parkinson’s Disease Rating Scale (UPDRS). However, these approaches are significantly influenced by subjectivity and exhibit insufficient sensitivity for early-stage symptom detection. The introduction of deep learning techniques has opened new avenues for the early diagnosis of PD. In contrast to traditional methods, deep learning models are capable of processing large-scale, high-dimensional, and complex datasets to automatically learn latent feature relationships, making them particularly suitable for scenarios involving multimodal data fusion. The multimodal diagnosis of PD is confronted with two enduring challenges: (1) the dependence on pre-existing knowledge of established genetic risk loci, and (2) the low efficiency and limited interpretability in handling interactions among cross-modal features. To address these challenges, this study introduces an innovative multimodal deep learning framework with two primary contributions: (1) a Genome-Wide Association Study (GWAS)-Transformer architecture that autonomously selects single nucleotide polymorphism (SNP) features through GWAS and utilizes a multi-head attention mechanism to model potential associations between non-risk loci, thereby eliminating the reliance on known susceptibility genes; (2) a Residual Attention Contrastive Fusion (RACF) module that tackles the heterogeneity of cross-modal features by dynamically allocating attention weights and applying contrastive loss constraints. Evaluation results on the Parkinson’s Progression Markers Initiative (PPMI) dataset demonstrate that our model achieves a classification accuracy of 91.2% and an AUC of 0.94, and predicts nine potential novel risk loci. This work presents a novel paradigm for the discovery of new risk loci based on deep learning and offers valuable insights from a multi-omics perspective for advancing PD research.

Improving the cleaning quality of tube lumen instruments by imaging analysis and deep learning techniques.

The complex structure of tube lumen instruments (TLIs) makes them more difficult to clean compared to solid instruments. This study aims to improve the cleaning quality inspection of reusable TLIs, ensuring patient safety and clinical reliability. This study improves the inspection of TLI cleaning quality using imaging analysis and deep learning techniques. Internally cleaned TLIs were imaged using an electronic endoscope by clinical staff, and the resulting images formed the original dataset. To enhance the quality of the TLI images and augment the dataset, image preprocessingtechniques such as enhancement, slicing, and threshold filtering were applied. Based on the sliced image dataset, baseline models with relatively better performance were selected by comparing the performance of multiple deep learning models in TLI image classification. To further improve the model's performance, two attention mechanisms were introduced to focus on important features. The optimized model outperforms the baseline model in both performance and stability. Specifically, the FA-ResNet18 model with the concurrent space and channelsqueeze and excitation (scSE) attention mechanism performs the best, with accuracy, macro precision, macro recall and macro F2 metrics all exceeding 98.3 %. This method can effectively reduce the risk of errors caused by subjective factors and visual fatigue in manual inspection.

Generation of Fake Mesh Structures Using GANs with SIFT-Based Feature Preservation

The mechanism used in processing geometric information for graphs lies at the intersection of graph theory with engineering and data science; in this case, the structured graph data is processed and analyzed through topological and geometric concepts, thus generating artificial graphs that are similar to real data, taking into account when creating the data, the preservation of the existing properties of the original graphs, which are the basic features and structural patterns. The proposed approach uses a generative adversarial network (GAN) with the addition of a completely separate layer for each of the generator and discriminator networks, which is one of the deep learning techniques; in this way, the system is enabled to identify the patterns present in real data and reduce the possibility of overfitting resulting during the processing of data which is in the form of vectors. Applying this strategy, the model can build fake graphs with the number of points equal to the number of input data points and whose location is close to the points present in the real data graphs. This study aims to produce graphs that are simulated to the original graphs. The simulated graph data can be used in the field of algorithm development, benchmarking and other fields such as testing and data augmentation.

Automated engineered-stone silicosis screening and staging using Deep Learning with X-rays.

Automated engineered-stone silicosis screening and staging using Deep Learning with X-rays.

Sound event detection by intermittency ratio criterium and source classification by deep learning techniques

Abstract Urban environments are characterized by a complex interplay of various sound sources, which significantly influence the overall soundscape quality. This study presents a methodology that combines the intermittency ratio (IR) metric for acoustic event detection with deep learning (DL) techniques for the classification of sound sources associated with these events. The aim is to provide an automated tool for detecting and categorizing polyphonic acoustic events, thereby enhancing our ability to assess and manage environmental noise. Using a dataset collected in the city center of Barcelona, our results demonstrate the effectiveness of the IR metric in successfully detecting events from diverse categories. Specifically, the IR captures the temporal variations of sound pressure levels due to significant noise events, enabling their detection but not providing information on the associated sound sources. To fill this weakness, the DL-based classification system, which uses a MobileNet convolutional neural network, shows promise in identifying foreground sound sources. Our findings highlight the potential of DL techniques to automate the classification of sound sources, providing valuable insights into the acoustic environment. The proposed methodology of combining the two above techniques represents a step forward in automating acoustic event detection and classification in urban soundscapes and providing important information to manage noise mitigation actions.

EEG-based schizophrenia detection: integrating discrete wavelet transform and deep learning.

Millions of people worldwide are afflicted with the psychological disease Schizophrenia (SZ). Symptoms of SZ include delusions, hallucinations, disoriented speech, and confused thinking. This disorder is manually diagnosed by a skilled medical practitioner. Nowadays, machine learning and deep learning techniques based on electroencephalogram (EEG) signals have been proposed to support medical practitioners. This paper proposes a deep learning system and a wavelet transform-based computer-aided detection method for detecting SZ disorder. The proposed technique aims to present a highly accurate EEG signal-based SZ detection technique. In this work, we first separate the EEG signal into sub-bands and extract the features for each sub-band using the Discrete Wavelet Transform (DWT). We have explored different mother wavelets and decomposition levels for the DWT setting; it is found that the Daubechies (db4) wavelet with 7-level decomposition performs the best for SZ detection. After obtaining the gathered features, the multilayer perceptron neural network (MLP) applies them to differentiate between SZ patients and healthy controls (HC). We validate our proposed automated SZ detection method using two publicly available datasets, Dataset-1 (DS1) with 81 records (32-HC and 49-SZ) and Dataset-2 (DS2) with 28 records (14-HC and 14-SZ), respectively. Compared with previous work, our proposed model surpasses the state-of-the-art technique for SZ detection. Our classification accuracy has increased, achieving an accuracy of 99.61% and 99.12% for DS1 and DS2. Our proposed method for identifying SZ using EEG signals is more reliable and accurate and is ready to support physicians in diagnosing SZ.

Leveraging Deep Learning for End-to-End Deepfake Video Identification

In recent years, the rise of deepfake technology has posed significant risks to media integrity, cybersecurity, and personal privacy. This project, Leveraging Deep Learning for End-to-End Deepfake Video Identification, proposes a robust detection system integrating spatial and temporal analysis with advanced deep learning techniques. The system employs a hybrid architecture, combining ResNet-50 for detecting frame-level artifacts and Long Short-Term Memory (LSTM) networks for capturing temporal inconsistencies across video frames. The methodology involves preprocessing video data, extracting frames, and applying normalization techniques to enhance feature clarity. ResNet-50 identifies anomalies like irregular facial textures and unnatural lighting, while LSTM tracks facial movements and blink rates. Additional techniques like Optical Flow Analysis, Discrete Fourier Transform (DFT), Principal Component Analysis (PCA), and Recursive Feature Elimination (RFE) further optimize the detection process. Performance evaluation is conducted using benchmark datasets with metrics such as accuracy, precision, recall, and AUC-ROC, demonstrating the system's robustness. A Flask-based web interface allows users to upload videos and obtain real-time feedback. Ethical considerations regarding privacy and responsible use are emphasized, promoting transparency and fairness. In conclusion, this project highlights the potential of combining deep learning and quantum computing paradigms to combat digital content manipulation.

A Linguistics-based deep learning Approach to ETL for Automated Translation of English Language Data

Automated translation of information regarding English language has become imperative in this worldwide phenomenon. The erstwhile methods of performing Extract, Transform, and Load (ETL) of data involve a lot of manual effort besides taking a long time and being error-prone. A linguistics-based deep learning approach is proposed to improve the efficiency and accuracy of ETL for automated translation of English language data. The particular approach is focused on the adoption of deep learning techniques for automatically processing and translating data in English language without extensive manual intervention. In conjunction with this, it provides the use of linguistics knowledge, such as syntax, semantics, and grammar, for building an accurate deep learning model for the extract-transform-load process of being applied in the translation process. The method has also proved to be promising in experimentation with results competing favorably with common ETL methods speed and accuracy and showing good scalability. Linguistic insights in combination with deep learning have created for the specified approach the possibility of bringing efficiency, accuracy, and automation to the translation of the English language.

Demystifying diagnosis: an efficient deep learning technique with explainable AI to improve breast cancer detection

As per a WHO survey conducted in 2023, more than 2.3 million breast cancer (BC) cases are reported every year. In nearly 95% of countries, the second leading cause of death for females is BC. Breast and cervical cancers cause 80% of reported deaths in middle-income countries. Early detection of breast cancer can help patients better manage their condition and increase their chances of survival. However, traditional AI models frequently conceal their decision-making processes and are mainly tailored for classification tasks. Our approach combines composite deep learning techniques with explainable artificial intelligence (XAI) to enhance interpretability and predictive accuracy. By utilizing XAI to examine features and provide insights into its classifications, the model clarifies the rationale behind its decisions, resulting in an understanding of concealed patterns linked to breast cancer detection. The XAI strengthens practitioners’ and health researchers’ confidence and understanding of artificial intelligence (AI)-based models. In this work, we introduce a hybrid deep learning bi-directional long short-term memory-convolutional neural network (BiLSTM-CNN) model to identify breast cancer using patient data effectively. We first balanced the dataset before using the BiLSTM-CNN model. The hybrid deep learning (DL) model presented here performed well in comparison to other studies, with 0.993 accuracy, precision 0.99, recall 0.99, and F1-score 0.99.

Advanced Software Techniques for Detecting Digital Image Manipulation

It is necessary to detect forgery of digital images in order to maintain their integrity. In this paper an attempt is made to solve the problem of copy-move forgery detection which is the most common form of image manipulation. We propose two new methods for detecting duplicated regions which are based on the texture and statistical features. The first technique is based on the classical SIFT (Scale-Invariant Feature Transform) which is a keypoint based method; the second one is based on the integration of SIFT into deep learning techniques and software solutions. We used the MICC-F600 database to evaluate the proposed methods and split it into training, validation and test sets. To increase performance of the model, some pre-processing steps were included such as scaling, image polishing, and so on. Based on the experimental results, our software-embedded convolutional neural network (CNN) models reached the greatest accuracy in identifying forged images. The model develops a binary mask that estimates the location of forgery in an image, which assists to improve the precision of digital forensic examination with our intelligent software.

Deep learning-based multi-criteria recommender system for technology-enhanced learning

Multi-Criteria Recommender Systems (MCRSs) improve personalization by incorporating multiple user preferences. However, their application in Technology-Enhanced Learning (TEL) remains limited due to challenges such as data sparsity, over-specialization, and cold-start problems. Traditional techniques, such as Singular Value Decomposition (SVD) and SVD + + , struggle to effectively model the complex interactions within multi-criteria rating data, leading to suboptimal recommendations. This paper introduces a hybrid DeepFM-SVD + + model, which integrates deep learning and factorization-based techniques to improve multi-criteria recommendations. The model captures both low-order feature interactions using factorization machines and high-order dependencies through deep neural networks, enabling more adaptive recommendations. To evaluate its performance, the model is tested on two multi-criteria datasets: ITM-Rec (TEL domain) and Yahoo Movies (non-TEL domain). The experimental results show that DeepFM-SVD + + consistently outperforms the traditional techniques across multiple evaluation metrics. The findings highlight significant improvements in accuracy, demonstrating the model’s effectiveness in sparse datasets and its generalization across domains. By addressing the limitations of existing MCRS techniques, this study contributes to advancing personalized learning recommendations in TEL and expands the applicability of deep learning-based MCRS beyond educational contexts.

Integrating DEM and Deep Learning for Forested Terrain Analysis: Enhancing Fire Risk Assessment Through Mountain Peak and Water System Extraction in Chongli District

Accurate fire risk assessment in forested terrain is crucial for effective disaster management and ecological conservation. This study innovatively proposes a novel framework that integrates Digital Elevation Models (DEMs) with deep learning techniques to enhance fire risk assessment in Chongli District. Our framework innovatively combines DEM data with Faster Regions with Convolutional Neural Networks (Faster R-CNN) and CNN-based methods, breaking through the limitations of traditional approaches that rely on manual feature extraction. It is capable of automatically identifying critical terrain features, such as mountain peaks and water systems, with higher accuracy and efficiency. DEMs provide high-resolution topographical information, which deep learning models leverage to accurately identify and delineate key geographical features. Our results show that the integration of DEMs and deep learning significantly improves the accuracy of fire risk assessment by offering detailed and precise terrain analysis, thereby providing more reliable inputs for fire behavior prediction. The extracted mountain peaks and water systems, as fundamental inputs for fire behavior prediction, enable more accurate predictions of fire spread and potential impact areas. This study not only highlights the great potential of combining geospatial data with advanced machine learning techniques but also offers a scalable and efficient solution for forest fire risk management in mountainous regions. Future work will focus on expanding the dataset to include more environmental variables and validating the model in different geographical areas to further enhance its robustness and applicability.

Prediction of Weld Geometry in Laser Overlap Welding of Low-Carbon Galvanized Steel

Accurate prediction of weld bead geometry is critical for optimizing laser overlap welding of low-carbon galvanized steel, as it directly affects joint quality and mechanical performance. Traditional finite element method (FEM)-based models provide reliable predictions but are computationally expensive and impractical for real-time applications. This study presents an artificial neural network (ANN)-based predictive model trained on a combination of experimental data and validated FEM simulations to estimate key weld characteristics, including depth of penetration (DOP), weld bead width at the surface (WS), and weld bead width at the interface (WI). The ANN model was evaluated using various improved statistical metrics. Results demonstrated a strong correlation between ANN predictions and experimental measurements, with R2 values exceeding 95% for WS and DOP and 92% for WI, and mean errors below 7%. A comparative analysis between ANN, FEM, and experimental data confirmed the model’s reliability across different welding conditions. Additionally, ANN significantly reduced computational time compared to FEM while maintaining high accuracy, making it a practical tool for real-time process optimization. These findings highlight the potential of ANN models as efficient alternatives to conventional simulation techniques in laser overlap welding applications. Future improvements may involve integrating real-time sensor data and deep learning techniques to further enhance predictive performance.