Browse Items (11809 total)
Sort by:
-
Hybrid Deep Learning-Based Potato andTomato Leaf Disease Classification
Predicting potato and tomato leaf disease is vital to global food security and economic stability. Potatoes and tomatoes are among the most important staple crops, providing essential nutrition to millions worldwide. However, many tomato and potato leaf diseases can seriously reduce food productivity and yields. We are proposing a hybrid deep learning model that combines optimized CNN (OCNN) and optimized LSTM (OLSTM). The weight values of LSTM and CNN models are optimized using the modified raindrop optimization (MRDO) algorithm and the modified shark smell optimization (MSSO) algorithm, respectively. The OCNN model is used to extract potato leaf image features and then fed into the OLSTM model, which handles data sequences and captures temporal dependencies from the extracted features. Precision, recall, F1-score, and accuracy metrics are used to analyze the output of the proposed OCNN-OLSTM model. The experimental performance is compared without optimizing the CNN-LSTM model, individual CNN and LSTM models, and existing MobileNet and ResNet50 models. The presented model results are compared with existing available work. We have received an accuracy of 99.25% potato and 99.31% for tomato. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Intelligent Approaches of Clinical and Nonclinical Type-1 Diabetes Data Clustering and Analysis
Every year in India, there are nearly 15,600 fresh cases being reported among these age groups. In 2011, in the United States, 18,000 children under 15 were newly reported for T1DM. Over 13years, the Karnataka state government has a list of records showing that out of 100,000, 37% of boys and 40% of girls are affected by T1DM Disease. This paper investigates two methodologies to identify significant details about Type-1 diabetes. The first methodology is applicable to clinical data. The second methodology is demonstrated for the NDA T1D dataset. The dataset is utilized further to apply machine learning techniques to group similar patient traits. Exploratory data analysis on the dataset has revealed significant information answering a few research questions. This analysis can be useful for India, China, and other countries with high populations. In this paper, a unique methodology based on Artificial Intelligence Technique is proposed for both clinical and non-clinical data. The Autoimmune Disease, Diabetes Type 1-T1D, is focused. Non Clinical data based on 2021 reports are collected to identify patterns. Substantial unique issues are addressed in this work which were never reported before. The knowledge generated can be helpful for creating new clinical datasets, methodology and new insights related to Type-1 diabetes. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Prevention and Mitigation of Intrusion Using an Efficient Ensemble Classification in Fog Computing
Cloud services in fog network is a platform that inherits software services to a network to handle cloud-specific problems. A significant component of the security paradigm that supports service quality is represented by intrusion detection systems (IDSs). This work develops an optimization environment to mitigate intrusion using RSLO classifier on a cloud-based fog networks. Here, a three-layer approach namely the cloud, end point, and fog layers is used as a trio to carry out all of the processing. In the cloud layer, three layers of processing are required for handling the dataset metrics which are data transformation metrics, feature selection metrics, and classification processes. With log transformation, data is transformed using KS correlation-based filter which is used to choose a feature. The classification using an ensemble methodology of RideNN classifiers which is a Rider Sea Lion Optimization (RSLO), a created classifier, is used to tune the ensemble classifier. Physical work is carried out at another layer called an end point layer. A trained ensemble classifier is used for intrusion detection in the fog layer. A greater precision, recall, and F-measure were obtained with an accuracy approximately 95%, with all benefits of the suggested RSLO-based ensemble strategy. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
EFMD-DCNN: Efficient Face Mask Detection Model in Street Camera Using Double CNN
The COVID-19 pandemic has necessitated the widespread use of masks, and in India, mask-wearing in public gatherings has become mandatory, with violators being fined. In densely populated nations like India, strict regulations must be established and enforced to mitigate the pandemics impact. Authorities and cameras conduct real-time monitoring of individuals leaving their homes, but 24/7 surveillance by humans is not feasible. A suggested approach to resolve this problem is to connect human intelligence and Artificial Intelligence (AI) by employing two Machine Learning (ML) models to recognize people who arent wearing masks in live-stream feeds from surveillance, street, and new IP mask recognition cameras. The effectiveness of this method has been demonstrated through its high accuracy compared to other algorithms. The first ML model uses the YOLO (You Only Look Once) model to recognize human faces in real-time video streams. The second ML model is a pre-trained classifier using 180,000 photos to categorize photos of humans into two groups: masked and unmasked. Double is a model that combines face recognition and mask classification into a single model. CNN provides a potential solution that may be utilized with image or video-capturing equipment such as CCTV cameras to monitor security breaches, encourage mask usage, and promote a secure workplace. This studys proposed mask detection technology utilized pre-trained datasets, face detection, and various classifiers to classify faces as having a proper mask, an improper mask, or no mask. The Double CNN-based model incorporated dual convolutional neural networks and a technology-based warning system to provide real-time facial identification detection. The ML model achieved high performance and accuracy of 98.15%, with the highest precision and recall, and can be used worldwide due to its cost-effectiveness. Overall, the proposed mask detection approach can potentially be a valuable instrument for preventing the spread of infectious diseases. The Author(s), under exclusive license to Springer Nature Switzerland AG 2024. -
Deep learning based federated learning scheme for decentralized blockchain
Blockchain has the characteristics of immutability and decentralization, and its combination with federated learning has become a hot topic in the field of artificial intelligence. At present, decentralized, federated learning has the problem of performance degradation caused by non-independent and identical training data distribution. To solve this problem, a calculation method for model similarity is proposed, and then a decentralized, federated learning strategy based on the similarity of the model is designed and tested using five federated learning tasks: CNN model training fashion-mnist dataset, alexnet model training cifar10 dataset, TextRnn model training thusnews dataset, Resnet18 model training SVHN dataset and LSTM model training sentiment140 dataset. The experimental results show that the designed strategy performs decentralized, federated learning under the nonindependent and identically distributed data of five tasks, and the accuracy rates are increased by 2.51, 5.16, 17.58, 2.46 and 5.23 percentage points, respectively. 2024 The Author(s). -
A Study of Simulated Working of A* and RRT* for Cargo Ship in ASVs
With the increased amount of algorithms for the path planning and collision avoidance of ASVs. The need for an unbiased protective path planning directs the need for decision in stochastic areas in the vast ocean for cargo ship. Autonomous surface vehicles should take appropriate decision on the path according to the dynamic environment and the obstacle that is before them. In some cases, environment, time, and size should be considered to acquire the fastest path and methods that could be suited for collision avoidance. This paper investigates the need for a well-known path planning method that has handled the situation based on the dynamic properties of the vehicle in the ocean. The simulated result shows a slight variation in their proposed path in terms of time and collision in terms of size. Therefore, using a realistic approach of the A* algorithm and the RRT*, we can handle the scenario of dynamic environment. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
A Review on Artificial Intelligence Techniques for Multilingual SMS Spam Detection
With social networks increased popularity and smartphone technology advancements, Facebook, Twitter, and short text messaging services (SMS) have gained popularity. The availability of these low cost text-based communication services has implicitly increased the intrusion of spam messages. These spam messages have started emerging as an important issue, especially to short-duration mobile users such as aged persons, children, and other less skilled users of mobile phones. Unknowingly or mistakenly clicking the hyperlinks in spam messages or subscribing to advertisements puts them under threat of debiting their money from either the bank account or the balance of the network subscriber. Different approaches have been attempted to detect spam messages in the last decade. Many mobile applications have also evolved for spam detection in English, but still, there is a lack of performance. As English has been completely covered under natural language processing, other regional languages, such as Urdu and Hindi variants, have specific issues detecting spam messages. Mobile users suffer greatly from these issues, especially in multilingual countries like India. Thus, this paper critically reviews the artificial intelligence-based spam detection system. The review lists out the existing systems that use machine and deep learning techniques with their limitations, merits, and demerits. In addition, this paper covers the scope for future enhancements in natural language processing to efficiently prevent spam messages rather than detect spam messages. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Artificial Neural Networks for Enhancing E-commerce: A Study on Improving Personalization, Recommendation, and Customer Experience
With e-commerce companies, artificial intelligence (AI) has emerged as a crucial innovation that allows companies to streamline processes, improve customer interactions, and increase operational capabilities. To provide tailored suggestions, address client care requests, and improve inventory control, AI systems may evaluate consumer data. Moreover, AI can improve pricing methods and identify fraudulent activity. Companies can actually compete and provide better consumer interactions with the growing usage of machine learning in e-commerce. This essay examines how AI is reshaping the e-commerce sector and creating fresh chances for companies to enhance their processes and spur expansion. AI technology which enables companies to enhance their procedures and offer a more individualized customer experiences has grown into a crucial component of the e-commerce sector. Purpose of providing product suggestions and improve pricing tactics, intelligent machines may examine consumer behavior, interests, and purchase history. Customer service employees will have less work to do as a result of chatbots powered by artificial intelligence handling client queries and grievances. AI may also aid online retailers in streamlining their inventory control by anticipating demands and avoiding overstocking. The use of AI technologies can also identify suspicious transactions and stop economic losses. AI is positioned to assume a greater part in the expansion and accomplishment of the e-commerce sector as it grows in popularity. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
The Road to Reducing Vehicle CO2 Emissions: A Comprehensive Data Analysis
In recent years, the influence of carbon dioxide (CO2) releases on the environment have become a major concern. Vehicles are one of the major sources of CO2 emissions, and their contribution to climate change cannot be ignored. This research paper aims to investigate the CO2 emissions of vehicles and compare them with different types of engines, fuel types, and vehicle models. The study was carried out by gathering information about the CO2 emissions of vehicles from the official open data website of the Canadian government. Data from a 7-year period are included in the dataset, which is a compiled version. There is a total of 220 cases and 9 variables. The data is analyzed using statistical methods and tests to identify the significant differences in CO2 emissions among different Car Models. The results indicate that vehicles with diesel engines emit higher levels of CO2 compared to those with gasoline engines. Electric vehicles, on the other hand, have zero CO2 emissions, making them the most environmentally friendly option. Furthermore, the study found that the CO2 emissions of vehicles vary depending on the type of fuel used. The study also reveals that the CO2 emissions of vehicles depend on the model and age of the vehicle. Newer models tend to emit lower levels of CO2 compared to older models. In conclusion, this study provides valuable insights into the CO2 emissions of Cars and highlights the need to adopt cleaner and more sustainable transportation options. The Author(s), under exclusive license to Springer Nature Switzerland AG 2024. -
Stock market prediction using DQN with DQNReg loss function
There have been many developments in predicting stock market prices usingreinforcement learning. Recently, Google released a paper that designed a new loss function,specifically for meta-learning reinforcement learning. In this paper, implementation is doneusing this loss function to the reinforcement learning model, whose objective is to predict thestock price based on certain parameters. The reinforcement learning used is an encoderdecoderframework that is useful for extracting features from long sequence prices. TheDQNReg loss function is implemented in the encoder-decoder model as it could providestrong adaptation performance in a variety of settings. The model can buy and sell the index, and the reward is the portfolio return after the days trading has concluded. To maximizeyield the model must optimize reward function. The DQNReg loss implemented DQN network and the Huber loss DQN network is compared with the Sharpe ratio considered for return. 2024 The Author(s). -
Detection of toxic comments over the internet using deep learning methods
People now share their ideas on a wide range of topics on social media, which has become an integral part of contemporary culture. The majority of people are increasingly turning to social media as a necessity, and there are numerous incidents of social media addiction that have been reported. Socialmedia channels. Socialmedia platforms have established their worth over time by bringing individuals from different backgrounds together, but they have also shown harmful side effects that could have serious consequences. One such unfavourable result is how extremely poisonous many discussions on social media are. Online abuse, hate speech, and occasionally outrage culture are now all considered to be toxic. In this study, we leverage the Transformers Bidirectional Encoder Representations to build an efficient model to detect and classify toxicity in user-generated content on social media. The Kaggle dataset with labelled toxic comments, was used to refine the BERT pre-trained model. Other Deep learning models, including Bidirectional LSTM, Bidirectional-LSTM with attention, and a few other models, were also tested to see which performed best in this classification task. We further evaluate the proposed models utilising dataset obtained from Twitter in order to find harmful content (tweets) using relevant hashtags. The findings showed how well the suggested methodology classified and analysed toxic comments. 2024 The Author(s). -
Comparative Analysis of State-of-the-Art Face Recognition Models: FaceNet, ArcFace, and OpenFace Using Image Classification Metrics
In recent years, facial recognition has emerged as a key technological advancement with numerous useful applications in numerous industries. FaceNet, ArcFace, and OpenFace are three widely used techniques for facial identification. In this study, we examined the accuracy, speed, and capacity to manage variations in face expression, illumination, and occlusion of these three approaches over a period of five years, from 2018 to 2023. According to our findings, FaceNet is more accurate than ArcFace and OpenFace, even under difficult circumstances like shifting lighting and facial occlusion. Also, during the previous five years, FaceNet has shown a significant improvement in performance. Even while ArcFace and OpenFace have made significant strides, they still lag behind FaceNet in terms of accuracy. Therefore, based on our findings, we conclude that FaceNet is the most effective method for facial recognition and is well-suited for use in high-stakes applications where accuracy is crucial. The Author(s), under exclusive license to Springer Nature Switzerland AG 2024. -
Advancing Collaborative AI Learning Through the Convergence of Blockchain Technology and Federated Learning
Artificial intelligence (AI) has revolutionized multiple sectors through its growth and diversification, notably with the concept of collaborative learning. Among these advancements, federated learning (FL) emerges as a significant decentralized learning approach; however, it is not without its issues. To address the challenges of trust and security in FL, this paper introduces a novel blockchain-based decentralized collaborative learning system and a decentralized asynchronous collaborative learning algorithm for the AI-based industrial Internet environment. We developed a chaincode middleware to bridge blockchain network and AI training for secure, trustworthy and efficient federated learning and presented a refined directed acyclic graph (DAG) consensus mechanism to reduce stale models impact, ensuring efficient learning. Our solutions effectiveness was demonstrated through application on an energy conversion prediction dataset from hydroelectric power generation, validating the practical applicability of our proposed system. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
A Unified Approach to Predict and Understand Acute Myeloid Leukemia Diagnosis
Acute myeloid leukemia (AML) is a rapidly progressing disease that affects myeloid cells in blood and bone marrow. These abnormal cancerous cells called blast cells are non-functional cells that increase rapidly in bone marrow and are released into blood stream which crowd out the healthy functional cells leading to weak immune system. This life-threatening disease needs to be diagnosed at early stage and hence requires fully automated system for early detection of leukemia to aid pathologists and doctors. Most of the automated machine learning and AI models are not transparent and require techniques to explain model prediction. This paper presents methods to classify blood microscopic images into healthy or acute myeloid leukemia. Among all the methods implemented, Gradient Boosting outperforms with an accuracy of 96.67%. This paper also focuses on explainable AI to interpret model prediction and feature importance which further helps in understanding decision-making process of classification model and optimize it. 2024, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. -
An Empirical and Statistical Analysis of Classification Algorithms Used in Heart Attack Forecasting
The risk of dying from a heart attack is high everywhere in the world. This is based on the fact that every forty seconds, someone dies from a myocardial infarction. In this paper, heart attack is predicted with the help of dataset sourced from UCI Machine Learning Repository. The dataset analyses 13 attributes of 303 patients. The categorization method of Data Mining helps predict if a person will have a heart attack based on how they live their lives. An empirical and statistical analysis of different classification methods like the Support Vector Machine (SVM) Algorithm, Random Forest (RF) Algorithm, K-Nearest Neighbour (KNN) Algorithm, Logistic Regression (LR) Algorithm, and Decision Tree (DT) Algorithm is used as classifiers for effective prediction of the disease. The research study showed classification accuracy of 90% using KNN Algorithm. The Author(s), under exclusive license to Springer Nature Switzerland AG 2024. -
Review and Design of Integrated Dashboard Model for Performance Measurements
This article presents a new approach for performance measurement in organizations, integrating the analytic hierarchy process (AHP) and objective matrix (OM) with the balanced scorecard (BSC) dashboard model. This comprehensive framework prioritizes strategic objectives, establishes performance measures, and provides visual representations of progress over time. A case study illustrates the methods effectiveness, offering a holistic view of organizational performance. The article contributes significantly to performance measurement and management, providing a practical and comprehensive assessment framework. Additionally, the project focuses on creating an intuitive dashboard for Fursa Foods Ltd. Using IoT technology, it delivers real-time insights into environmental variables affecting rice processing. The dashboard allows data storage, graphical representations, and other visualizations using Python, enhancing production oversight for the company. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Arabica Coffee Bean Grading into Specialty and Commodity Type Based on Quality Using Visual Inspection
Expanding potential of coffee consumers to seek out the freshest and best flavors is a cause for the rise of specialty coffee inthe market. Specialty coffee is grown and harvestedmaintaining an emphasis on quality and clarity of flavor, whereas commodity coffee is harvested for caffeine content. Within those inclusive categories, arabica and robusta are the two types of main branches of coffee that weencounter in the coffee market. Specialty coffees differ significantly from conventional coffees in that they are cultivated at higher altitudes, can be traced, and are professionally processed after being harvested. The quality is constantly examined and understood at every stage, from growth to brewing. Green arabica quality is assessed by counting the defective beans present in the sample. These defects can be primary (Category I) or secondary (Category II). If the primary defects are null and less than five secondary defects, coffee is said to be a specialty.Prior research has been done on classifying the coffee species and differentiating good beans from bad beans. Our research involves the combination of machine learning like K-NN and deep learning convolutional neural networks for classifying specialty coffee from commodity type using computer vision. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Multimodal Face and Ear Recognition Using Feature Level and Score Level Fusion Approach
Recent years have seen a significant increase in attention in multimodal biometric systems for personal identification especially in unconstrained environments. This paper presents a multimodal recognition system by combining feature level fusion of ear and profile face images. Multimodal biometric systems by combining face and ear can be used in an extensive range of applications because we can capture both the biometrics in a non-intrusive manner. Local texture feature descriptor, BSIF is used to extract discriminative features from biometric templates. Feature level and score level fusion is experimented to improve the performance of the system. Experimental results on different public datasets like GTAV, FEI, etc., show that the proposed method gives better performance in recognition results than individual modality. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Improving Speaker Gender Detection by Combining Pitch and SDC
Gender detection is helpful in various applications, such as speaker and emotion recognition, which helps with online learning, telecom caller identification, etc. This process is also used in speech analysis and initiating human-machine interaction. Gender detection is a complex process but an essential part of the digital world dealing with voice. The proposed approach is to detect gender from a speech by combining acoustic features like shifted delta cepstral (SDC) and pitch. The first step is preprocessing the speech sample to retrieve valid speech data. The second step is to calculate the pitch and SDC for each frame. The multifeature fusion method combines the speech features, and the XGBoost model is applied to detect gender. This approach results in accuracy rates of 99.44 and 99.37% with the help of RAVDESS and TIMIT datasets compared to the pre-defined methods. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024. -
Forex Analysis on USD to INR Conversion: A Comparative Analysis of Multiple Statistical and Machine Learning Algorithms
Foreign Currency Exchange (FOREX) engages a major role in world economy and the international market. It is a vast study based on determining whether or not to wait, buy or sell on a trading currency pair. The main objective is to predict the future currency prices using historical data in order to make more informed and accurate investment decisions for business traders and monetary market. This work experimented and implements ten machine learning strategies namely Random Forest, Decision Tree, Support vector regressor (SVM), Linear SVM, Linear Regression, Ridge, Lasso, K-Nearest Neighbor (KNN), Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) to assess the historical data and help the traders to invest in foreign currency exchange. The dataset used to validate and verify the machine learning algorithms is available in public domain and it is the daily Foreign Currency Exchange price of United States Dollars (USD) to Indian Rupees (INR). The experimented result shows that the Long Short-Term Memory (LSTM) model performs a bit better than the other machine learning models for this particular case. This work straight away does not reject the other methods it rather needs more experimental analysis with other models that has changed architecture and different dataset. 2024 IEEE.