IAES International Journal of Artificial Intelligence (IJ-AI)
Not a member yet
1769 research outputs found
Sort by
Predicting the severity of road traffic accidents Morocco: a supervised machine learning approach
Get PDFEarly prediction of road accidents fatality and injuries severity is one of the important subjects to road safety emphasizing the critical need to prevent serious consequences to reduce injuries and fatalities. This study uses real road accidents data set in Morocco. It represents the intersection between road safety and data science, aiming to employ machine learning techniques to provide valuable insights in accident’s severity prevention. The purpose of this paper is to study road accidents data in the country and combine results from statistical methods, spatial analysis, and machine learning models to determine which factors will mostly contribute to increase the accident’ severity in the country. A comparison of results obtained was also conducted in this paper using different metrics to evaluate the effectiveness of each method and determine the most important factors that contribute to increase the fatality or injuries severity in the specific context of accidents. The best prediction model was then injected into a proposed algorithm where more intelligent techniques are included to be implemented in a car engine to perform an early detection of severe accidents and therefore preventing crashes from happening
Optimizing battery life: a TinyML approach to lithium-ion battery health monitoring
Get PDFElectrical vehicles (EVs) are crucial nowadays due to their reduction in greenhouse gas emissions, decreasing dependence on remnant fuels, and improving air quality. For EVs, the battery is the heart that determines range, performance, and efficiency. Also, it directly impacts the cost and overall vehicle life span. Lithium-ion (Li-ion) batteries are pivotal in powering modern portable electronics and electric vehicles due to their high energy density and durability. Issues with current batteries include slow charging, short cycles, and low energy density. Most of the problems with current batteries are resolved by Li-ion batteries, which also helps explain why EV usage is increasing globally. However, to guarantee maximum performance and safety, estimating the remaining useful life and health state of these batteries remains a major difficulty. To improve battery lifetime of the battery and to overcome the problems of delayed charging, this study introduces a tiny machine learning (TinyML) method. An innovative machine learning approach is put forth that allows for effective learning on devices with limited resources, which enables real-time monitoring of the health status of the Li-ion batteries
The use of geographic information systems to measure the financial performance of micro enterprises
Get PDFThis study examines the application of geographic information systems (GIS) to measure and visualize the financial performance of micro enterprises in remote areas of East Kalimantan, Indonesia. Micro enterprises are crucial to local economies but often face barriers such as limited capital access, inadequate infrastructure, and insufficient business training. Using a mixed-method approach, the research combined surveys of 200 micro business owners, secondary economic data, and GIS-based spatial analysis. The results indicate clear spatial disparities: enterprises located closer to financial institutions and training programs achieved 25–30% higher profitability and stronger operational resilience. GIS mapping effectively identified performance clusters and underserved zones, providing actionable insights for targeted policy interventions. Key factors influencing financial outcomes include access to capital, training opportunities, and infrastructure quality. This study demonstrates the value of GIS as a decision-support tool for policymakers in designing spatially informed financial assistance, infrastructure planning, and mobile training deployment. The findings contribute to socio-economic planning discourse and propose a replicable GIS-based framework for strengthening microenterprise resilience in underdeveloped regions
Optimizing brain tumor MRI classification using advanced preprocessing techniques and ensemble learning methods
Get PDFBrain tumor classification is a critical task in medical imaging that directly impacts the accuracy of diagnosis and treatment planning. However, the complexity and variability of magnetic resonance imaging (MRI) images pose significant challenges, often resulting in reduced model reliability and generalization. This study addresses these limitations by proposing a novel ResNet+Bagging model, leveraging the strengths of residual networks and ensemble learning to enhance classification performance. Using publicly available brain tumor MRI datasets, including images labeled as benign, malignant, and normal, the study employs advanced preprocessing techniques such as normalization, data augmentation, and noise reduction to ensure high-quality inputs. The proposed model demonstrated significant improvements, achieving the highest testing accuracy of 72%, outperforming other tested models such as LeNet, standard ResNet, GoogleNet, and VGGNet. Precision (0.6010), recall (0.6000), and F1-score (0.5990) metrics further highlight its superior balance in detecting positive and negative classes. The novelty of this research lies in the application of Bagging to ResNet, which effectively mitigates overfitting and enhances predictive stability in complex medical datasets. These findings underscore the proposed model's potential as a robust solution for brain tumor classification, contributing to more accurate and reliable diagnostics
Image segmentation using fuzzy clustering for industrial applications
Get PDFThis paper presents a fuzzy logic clustering algorithm oriented to image segmentation and the procedure designed to evaluate its performance by varying two parameters: the number of clusters (c) and the diffusivity parameter (m), which leads to the conclusion that an adjusted number of clusters is sufficient to recognize main elements of the image, but a more detailed reconstruction requires a higher number of clusters. Also, the diffusivity parameter influences the smoothness of the boundaries between clusters, low values generate a segmentation with more abrupt transitions and sharper contours, high values smooth the segmentation, its excessive increase may cause the elements to merge, losing details. In general, the balance between these two parameters is key to obtaining an effective segmentation. Three validation scenarios were used, the first two allowed to establish the most appropriate parameters for segmentation, regulating the clusters to a maximum of 4 and keeping the diffusivity level at 2.0, the third scenario validated the algorithm with real images of industrial cleaning products, all with noise, establishing the computational cost and processing times for images of 350×350 and 2000×3000 pixels resolution. In conclusion, applications of the algorithm are foreseen in automatic quality control and inventory control of finished products and raw materials, thanks to its high efficiency and low response time, even in scenarios involving noisy and large images
Enhancing waste management through municipal solid waste classification: a convolutional neural network approach
Get PDFThe escalation of population, economic expansion, and industrialization has resulted in an increase in waste production. This has made waste management more challenging and has resulted in environmental deterioration, negatively impacting the quality of life. Recycling, reducing, and reusing are viable methods to eradicate the escalating waste issue, requiring the appropriate classification of municipal solid waste. This study focuses on comparing six advanced waste classification systems that employ a pre-trained convolutional neural network (CNN) designed to recognize twelve distinct categories of municipal waste. It has been determined that DarkNet53 is the most effective classifier among these six models. To assess the effectiveness of each waste classifier, the confusion matrix, precision, recall, F1 score, the area under the receiver operating characteristic curve, and the loss function are examined. It has been found that DarkNet53 has an F1 score of 98.7% and validation accuracy of 99%, respectively. The suggested approach will be useful in promoting garbage recovery and reuse in the direction of a circular and sustainable economy
Machine and deep learning classifiers for binary and multi-class network intrusion detection systems
Get PDFThe rapid proliferation of the internet and advancements in communication technologies have significantly improved networking and increased data vol ume. This phenomenon has subsequently caused a multitude of novel attacks, thereby presenting significant challenges for network security in the intrusion detection system (IDS). Moreover, the ongoing threat from authorized entities who try to carry out various types of attacks on the network is a concern that must be handled seriously. IDS are used to provide network availability, confidentiality, and integrity by employing machine learning (ML) and deep learning (DL) algorithms. This research aimed to study the impacts of the binary and multi-attack instances label by establishing IDS that leverages hybrid algorithms, including artificial neural networks (ANN), random forest (RF), and logistic model trees (LMTs). The paper addresses challenges such as data pre processing, feature selection, and managing imbalanced datasets by applying synthetic minority oversampling technique (SMOTE) and Pearson’s correlation methodologies. The IDS was tested using network security laboratory knowledge discovery datasets (NSL-KDD) and catalonia independence corpus intrusion detection system (CIC-IDS-2017) datasets, achieving an average F1-score of 96% for binary classification on NSL-KDD and 85% for binary classification on CIC-IDS-2017, while for multi-classification, the proposed model achieved an average F1-score of 82% and 96% for NSL-KDD and CIC-IDS-2017 successively
An adaptive window function based on enhanced cuckoo search optimization for finite impulse response filter design
Get PDFThis study introduced a modern approach involving an adaptive window function with the enhanced cuckoo search optimization (ECSO) algorithm for optimizing the finite impulse response (FIR) filter design by dynamically adjusting window parameters. This proposed method enhanced spectral performance, and improved accuracy, resolution, and reliability in spectral analysis. A mathematical model was developed for the adaptive window function, and the original cuckoo search optimization (CSO) algorithm was enhanced through adaptive step-size adjustment. Results demonstrated better spectral characteristics with narrower main lobes, lower sidelobes, and enhanced stopband attenuation, indicating computational efficiency, versatility, and robustness. Comparative analysis showed that the adaptive window function outperformed Kaiser, Gaussian, Tukey, and Chebyshev windows, exhibiting superior frequency selectivity, uniform amplitude response within the passband, and improved signal fidelity with reduced interference from neighboring frequency bands. Additionally, it demonstrated lower leakage factors, indicating reduced spectral leakage and better confinement of signal energy within the desired frequency range. This advancement in FIR filter design holds promise for various signal processing tasks and real-time applications, marking a significant milestone in signal processing innovation
Heterogeneous semantic graph embedding assisted edge-sensitive learning for cross-domain recommendation
Get PDFIn the digital age, recommendation systems navigate vast alternatives. Content-based, collaborative filtering, deep-driven, and cross-domain recommendation (CDR) have been studied significantly but face cold-start and data sparsity. Though CDR methods outperform others, they struggle to optimize user-item matrices. Recent graph-based CDR methods improve efficiency by leveraging additional user-item interactions; however, optimizing graph features remains an open research area. Moreover, current techniques do not consider the impact of noise items (unrelated) on recommendation accuracy. To address this gap, this paper develops a heterogeneous semantic graph-embedding (HSGE) edge-pruning model that leverages user ratings and item metadata in the source and target domains to recommend items to target domain users. To achieve it, at first Word2Vec method is applied to explicit and implicit details, followed by Node2Vec-driven graph embedding matrix generation. Our HSGE method obtains user-user, user-item, and item-item connections to achieve more semantic features. To improve accuracy, our model prunes edges that drop source domain items and allied edges unrelated to the target domain users. Subsequently, the retained HSGE matrices from both domains are processed for element-wise attention. A multi-layer perceptron with cosine similarity processed combined features matrices to generate top-N recommendations with superior hit-rate (HR) and normalized discounted cumulative gain (NDCG)
MedicalBERT: enhancing biomedical natural language processing using pretrained BERT-based model
Get PDFRecent advances in natural language processing (NLP) have been driven by pretrained language models like BERT, RoBERTa, T5, and GPT. These models excel at understanding complex texts, but biomedical literature, with its domain-specific terminology, poses challenges that models like Word2Vec and bidirectional long short-term memory (Bi-LSTM) can't fully address. GPT and T5, despite capturing context, fall short in tasks needing bidirectional understanding, unlike BERT. Addressing this, we proposed MedicalBERT, a pretrained BERT model trained on a large biomedical dataset and equipped with domain-specific vocabulary that enhances the comprehension of biomedical terminology. MedicalBERT model is further optimized and fine-tuned to address diverse tasks, including named entity recognition, relation extraction, question answering, sentence similarity, and document classification. Performance metrics such as the F1-score, accuracy, and Pearson correlation are employed to showcase the efficiency of our model in comparison to other BERT-based models such as BioBERT, SciBERT, and ClinicalBERT. MedicalBERT outperforms these models on most of the benchmarks, and surpasses the general-purpose BERT model by 5.67% on average across all the tasks evaluated respectively. This work also underscores the potential of leveraging pretrained BERT models for medical NLP tasks, demonstrating the effectiveness of transfer learning techniques in capturing domain-specific information