International Journal of Research in Computing

A Comprehensive Review: Enhance Logistics Performance by Optimizing Supply Chain Routes with Dynamic Factors using Genetic Algorithm

2025-01-20T16:02:08+00:00

As supply chain networks grow increasingly complex, achieving optimal logistics has become essential for industries to remain competitive and adapt to dynamic demands. Traditional route optimization methods often fail to accommodate real-time factors such as traffic congestion, unpredictable weather conditions, and shifting customer requirements, leading to inefficiencies in logistics performance. This study aims to address these challenges by exploring the potential of Genetic Algorithm (GA) as a robust solution for multi-objective route optimization. A thematic literature review was conducted to evaluate existing algorithms and models, revealing significant gaps in their ability to manage dynamic, multi-factor logistics environments effectively. The review identified that Genetic Algorithm excel in integrating real-time data, enabling the optimization of delivery routes with greater efficiency and adaptability. Real-world applications of GA in diverse industries demonstrated reductions in delivery times, improved resource utilization, and enhanced customer satisfaction. These findings establish GA as an intelligent and scalable approach to modern logistics challenges, offering significant implications for advancing supply chain management practices.

Deep Learning Approaches for Classifying Informal and Formal English Texts Using Linguistic Features

2024-09-18T16:28:05+00:00

Effective techniques for automatically classifying texts are becoming increasingly necessary due to the exponential expansion of digital material. Differentiating between formal and informal documents can help students identify appropriate resources for their assignments and improve the effectiveness of information retrieval systems. Although machine learning is extensively utilized in classification of text, there is a lack of research focused to the effective differentiation of formal and informal writings through linguistic features. This gap highlights the necessity for advanced methodologies that improve classification accuracy and enhance the value of digital content in academic and retrieval systems. Our research addresses the problem by utilizing deep learning methodologies and a wide range of 13 linguistic attributes to get enhanced efficacy in text classification. Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long Short Term Memory Networks (LSTM) were considered. A dataset , including both formal (news articles, formal documents) and informal (personal letters, personal blogs) texts, were gathered from several web sources. We considered linguistic markers such as colloquialisms, contractions, modal verbs, slang, acronyms, pronouns, phrasal verbs, grammar complexity, vocabulary complexity, voice, and language type to generate the feature vector. The feature vectors were utilized to train and assess the classification models using several cross validation techniques, particularly 3, 5, 7, and 10 folds. The efficacy of the models was evaluated using performance indicators, f-measure, accuracy, precision, and recall. With the highest accuracy of 99.8% and resilience in differentiating between formal and informal texts, the LSTM model outperformed than the others. Future research will examine big datasets, more linguistic characteristics, sophisticated deep learning models, and real-time and multilingual classification systems.

Development of a Web App for Asthmatic Wheeze Detection using Convolutional Neural Networks

2024-12-19T05:40:37+00:00

Asthma and Chronic Obstructive Pulmonary Disease (COPD) are critical lung conditions characterized by breathing difficulties. In asthma, airways become constricted, inflamed, and filled with mucus, leading to symptoms such as wheezing, coughing, and shortness of breath. Wheezing serves as a vital diagnostic indicator for these and other respiratory disorders. Early detection and management are crucial to prevent severe complications and improve patient outcomes. This research introduces a web application for asthmatic wheeze detection, employing Convolutional Neural Networks (CNNs) to enable early identification of respiratory disorders in Sri Lanka. Our system captures audio recordings from an electronic stethoscope, processes the data using a CNN model, and detects wheezes with an impressive accuracy of 84%. The application not only identifies wheezing but also provides tailored therapy recommendations and dosage prescriptions based on the detected condition. By leveraging this advanced technology, we aim to revolutionize respiratory health monitoring in Sri Lanka, offering healthcare professionals a reliable tool for timely intervention and enhancing patient care.

Conversational AI for Cinnamon and Coffee Exports: Insights on Price and Yield

2024-11-28T16:12:15+00:00

This research covers the development of an AI-powered chatbot that will help develop the agricultural industry in Sri Lanka by answering queries regarding coffee and cinnamon, besides giving weekly producer’s price predictions for them. It uses an SVM classifier that selects suitable responses from a given query in Sinhala, translates into English, generates the response, and then translates back to Sinhala for presentation. It implements an LSTM model to forecast prices of export crops from 2016 to 2022. It was observed that there is a great correlation between crop prices and the start date of the week they are valid, with a Pearson coefficient of over 0.70 for both coffee and cinnamon, while others are below 0.60. The chatbot returned to an accuracy rate of 70% in the classification of queries, while poor performance was obtained for harvest prediction due to a lack of sufficient data. The successful integration of predictive models and the chatbot proves the potential of AI in improving agricultural decision-making, productivity, and efficiency. This research consists of a Sinhala language-based chatbot, providing customized advisory services and weekly price predictions, contributing to localized technological advancements in Sri Lankan agriculture.

An Image-Based Facial Emotion Detection Chatbot

2024-11-28T16:10:10+00:00

In the evolving domain of conversational AI, integrating visual recognition capabilities into chatbots represents a pivotal step toward achieving empathetic and context-aware interactions. This study introduces an innovative emotion-aware chatbot system that utilizes facial emotion recognition (FER) to enhance emotional intelligence in human AI communication. The primary problem addressed is the lack of conversational systems capable of interpreting non-verbal cues, such as facial emotions, to create meaningful and personalized interactions. Our chatbot allows users to input facial images, enabling the system to recognize and classify emotions in real-time and dynamically generate emotion-based responses tailored to the user's state. The FER model was developed using the FER-2013 benchmark dataset, categorizing expressions into seven predefined emotions: Angry, Disgust, Fear, Happy, Sad, Surprise, and Neutral. To address achieved moderate results, data augmentation techniques and hyperparameter tuning were applied to improve robustness. Furthermore, LangChain, an open-source framework for building conversational agents, was integrated to manage dialogue flows. LangChain was utilized to orchestrate the chatbot’s conversational flow, leveraging its modular architecture for dynamic and adaptive dialogue management textually and visually. Recognized emotions from the FER model were processed by LangChain to generate contextually relevant responses tailored to the user's emotional state. The framework enabled seamless integration of visual input processing with language-based conversation, ensuring smooth transitions between emotion recognition and response generation. The integration methodology leverages LangChain’s toolkits for real-time processing of visual cues, enabling emotion-driven, contextually adaptive conversation generation. Unlike conventional chatbots, this system introduces a multimodal approach that bridges textual and visual emotional inputs with the integration of LangChain. This research contributes a detailed framework for integrating FER into conversational agents, emphasizing its potential in building rapport, improving engagement, and creating empathetic dialogue. Future work will focus on optimizing the FER model’s accuracy through advanced architectures and exploring real-world use cases, including healthcare and customer service, to demonstrate the transformative impact of emotion-aware AI on communication platforms. Future work will focus on improving FER model performance through advanced architectures like Vision Transformers and larger, more diverse datasets to boost accuracy and generalizability.

Faces Unveiled: A Deep Dive into Modern Face Detection and Recognition Techniques

2024-09-05T02:08:31+00:00

This paper provides a comprehensive overview of contemporary research in face detection, facial feature detection, and face recognition, categorizing methodologies into four primary types: knowledge-based, template matching, feature based, and appearance-based. Analysis reveals a predominant focus on appearance-based techniques, particularly in recent studies. Literature showcases the increasing utilization of deep learning algorithms, such as CNN, DCNN, and Faster RCNN, to address challenges in face detection and recognition. Notably, these algorithms demonstrate high accuracy in complex scenarios, including variations in pose, scale, and occlusion. The overview highlights the effectiveness of knowledge-based methods in detecting facial features with low computational requirements, albeit with limited accuracy in complex situations. Appearance-based methods, particularly those employing deep learning, emerge as highly successful in face detection and recognition, achieving accuracy rates exceeding 99%. The integration of one stage and two-stage algorithms, coupled with traditional classifiers, underscores their efficacy. Researchers enhance accuracy through data augmentation, multi-task learning, and network acceleration techniques. The paper concludes that deep learning algorithms significantly impact face detection, recognition, and feature extraction, reflecting their pivotal role in advancing computer vision. The comprehensive review of 28 selected papers emphasizes the importance of continued research to further enhance these essential aspects of object detection.