International Journal of Informatics and Communication Technology (IJ-ICT)
Not a member yet
494 research outputs found
Sort by
Leveraging distillation token and weaker teacher model to improve DeiT transfer learning capability
Get PDFRecently, distilling knowledge from convolutional neural networks (CNN) has positively impacted the data-efficient image transformer (DeiT) model. Due to the distillation token, this method is capable of boosting DeiT performance and helping DeiT to learn faster. Unfortunately, a distillation procedure with that token has not yet been implemented in the DeiT for transfer learning to the downstream dataset. This study proposes implementing a distillation procedure based on a distillation token for transfer learning. It boosts DeiT performance on downstream datasets. For example, our proposed method improves the DeiT B 16 model performance by 1.75% on the OxfordIIIT-Pets dataset. Furthermore, we present using a weaker model as a teacher of the DeiT. It could reduce the transfer learning process of the teacher model without reducing the DeiT performance too much. For example, DeiT B 16 model performance decreased by only 0.42% on Oxford 102 Flowers with EfficientNet V2S compared to RegNet Y 16GF. In contrast, in several cases, the DeiT B 16 model performance could improve with a weaker teacher model. For example, DeiT B 16 model performance improved by 1.06% on the OxfordIIIT-Pets dataset with EfficientNet V2S compared to RegNet Y 16GF as a teacher model
A decision support system for mushroom classification using Naïve Bayesian algorithm
Get PDFMushrooms are rich in vitamins and proteins, a well-known superfood, however, cases of harmful mushroom consumption worldwide result in hallucinations, illness, or death. A significant challenge is that some poisonous mushrooms closely resemble edible varieties, making it difficult for mushroom foragers to distinguish between them. This study introduced KabuTeach, a decision support system (DSS) designed to classify mushrooms based on their morphological characteristics using the Naïve Bayes (NB) algorithm. The classification model was applied to a real-world dataset of 8,124 instances from Kaggle, containing 23 attributes. Evaluation metrics, including accuracy, recall, precision, specificity, and F1-score, were used to assess the classifier’s performance. Results indicated that the NB classification algorithm integrated into KabuTeach achieved a high accuracy level of 89.13%, using a 70:30 data split and 5-fold cross-validation approaches. The 0.98 AUC (area under the curve) value further concluded that the model was excellent in classifying between edible and poisonous mushrooms. These findings showed that KabuTeach is a reliable classification tool that aids mushroom foragers in differentiating mushrooms and promoting safer consumption practices. This innovation in agricultural technology could potentially reduce health risks by minimizing accidental ingestion of toxic mushrooms, ultimately contributing to public health safety
Fuzzy logic-based driver fatigue prediction system for safe and eco-friendly driving
Get PDFThe advancement of intelligent car systems in recent years has been significantly influenced by developments in information technology. Driver fatigue is a dominant problem in car accidents. The goal of advanced driving assistance is to develop an advanced driving assistance system (ADAS) a eco-friendly model which focuses on the detection of drowsy driver, to notify drivers of their fatigued condition to prevent accidents on the roads. With relation to driving, the driver mustn’t be distracted by alarms when they are not tired. The answer to this unanswered question is provided by 60- second photograph sequences that were taken when the subject’s face was visible. To reduce false positives, two alternative solutions for determining whether the driver is drowsy have been developed. To extract numerical data from photos and feed it into a fuzzy logic-based system, convolutional network is applied initially; later deep learning technique is followed. The fuzzy logic-based solution avoids the false alarm of the system
A high linearity low noise amplifier with modified differential inductor for bluetooth profiles
Get PDFIn today’s rapidly evolving communication landscape, electronic devices rely heavily on high-performance components to ensure seamless connectivity. A low-noise amplifier (LNA) is a critical front-end element in any receiver chain, where its performance significantly influences the overall system efficiency. As integrated circuits continue to shrink with advancements in technology, challenges such as linearity degradation have become increasingly prominent. This work presents a modified derivative (MD) narrowband common source low-noise amplifier (CSLNA) designed using 0.13 µm CMOS technology, offering improved linearity and frequency characteristics. The proposed design adopts a hybrid architecture, combining a folded cascode gain stage with a common-gate configuration. An optimized modified differential inductor is employed at the input for effective impedance matching and reduced noise figure (NF). The implemented LNA achieves a gain of 25.81 dB, an input return loss of –24.86 dB, and maintains a low NF of 0.3 dB at an operating frequency of 2.4 GHz. Furthermore, the linearity metrics-third-order input intercept point (IIP3) and 1 dB compression pointare significantly improved to –16.70 dBm and –21.89 dBm, respectively. These results highlight the LNA's suitability for Bluetooth and other shortrange wireless communication applications
Renewable energy optimization for sustainable power generation
Get PDFTo improve sustainability in power generation, this study presents a thorough data-driven method for maximizing renewable energy sources. It employs measures like capacity utilization factor (CUF) and efficiency to evaluate the performance of solar and wind energy using historical weather and energy-generating data. The study offers practical suggestions for improving renewable energy systems, such as weather-energy correlation analysis and machine learning-based forecasting models. In addition, a comparative analysis is carried out to ascertain which energy source is better, and useful real-world data is provided, including a summary of all India’s total renewable energy generation (excluding large hydro) for June 2023 and a performance comparison year over year. A useful, data-driven approach for enhancing renewable energy is provided by this work, which advances the topic of sustainable energy
A comparative analysis of PoS tagging tools for Hindi and Marathi
Get PDFMany tools exist for performing parts of speech (PoS) data tagging in Hindi and Marathi. Still, no standard benchmark or performance evaluation data exists for these tools to help researchers choose the best according to their needs. This paper presents a performance comparison of different PoS taggers and widely available trained models for these two languages. We used different granularity data sets to compare the performance and precision of these tools with the Stanford PoS tagger. Since the tag sets used by these PoS taggers differ, we propose a mapping between different PoS tagsets to address this inherent challenge in tagger comparison. We tested our proposed PoS tag mappings on newly created Hindi and Marathi movie scripts and subtitle datasets since movie scripts are different in how they are formatted and structured. We shall be surveying and comparing five parts of speech taggers viz. IMLT Hindi rules-based PoS tagger, LTRC IIIT Hindi PoS tagger, CDAC Hindi PoS tagger, LTRC Marathi PoS tagger, CDAC Marathi PoS tagger. It would also help us evaluate how the Bureau of Indian Standards’s (BIS) tag set of Indian languages compares to the Universal Dependency (UD) PoS tag set, as no studies have been conducted before to evaluate this aspect
Classification and regression tree model for diabetes prediction
Get PDFDiabetes mellitus is characterized by excessive blood glucose that occurs when the pancreas malfunctions while producing insulin. High blood glucose levels can cause chronic damage to organs, particularly the eyes and kidneys. Diabetes prediction models traditionally use a variety of machine learning (ML) algorithms by combining data from the glucose levels, patient health parameters, and other biomarkers. Prior research on diabetes prediction using various algorithms, such as support vector machine (SVM) and decision tree (DT) models, demonstrates an accuracy rate of approximately 70%, which is relatively modest. Therefore, in this study, a classification and regression tree (CART) multiclassifier model has been proposed to improve the accuracy of diabetes prediction, which is based on three classes: non-diabetic, pre-diabetic, and diabetic. The study involved data preprocessing steps, hyperparameter tuning, and evaluation of performance metrics. The model achieved 97% accuracy while utilizing the value of 5 for the number of leaves per node, the value of 10 for the maximum number of splits, and deviance as the split criterion, which also resulted in a precision of 98%, recall of 97%, and F1-score of 98%, showing that the proposed multiclassifier model can accurately predict diabetes. In conclusion, the proposed CART model with the best hyperparameter setting can enable the highest accuracy in predicting diabetes classes
Adaptive intrusion detection system with DBSCAN to enhance banking cybersecurity
Get PDFThe accelerating pace of digital transformation in the banking sector has highlighted the critical need for comprehensive cybersecurity strategies capable of countering evolving cyber threats. This study introduces an innovative intrusion detection framework tailored for banking environments, leveraging the CICIDS2017 and CSECICIDS2018 datasets for evaluation and validation. The proposed framework integrates data preprocessing, feature reduction, and advanced attack detection methods to enhance detection accuracy. A basic autoencoder is utilized for dimensionality reduction, streamlining input data while preserving essential attributes. The density-based spatial clustering of applications with noise (DBSCAN) algorithm is then applied for attack detection, enabling the detection of intricate attack patterns and their classification into specific attack groups. The proposed adaptive intrusion detection system (IDS) framework demonstrates outstanding performance, achieving precision, recall, F1-score, and accuracy rates exceeding 98%. Comparative evaluations against conventional techniques, such as support vector machines (SVM), long short-term memory (LSTM), and K-means, highlight its superiority in terms accuracy and computational efficiency. This research address key challenges, including high-dimensional datasets, class imbalance, and dynamic threat landscapes, offering a scalable and efficient solution to enhance the security of banking operations and enable proactive threat mitigation in the sector
Enhanced smart farming security with class-aware intrusion detection in fog environment
Get PDFThe adoption of the internet of things (IoT) in smart farming has enabled real-time data collection and analysis, leading to significant improvements in productivity and quality. However, incorporating diverse sensors across large-scale IoT systems creates notable security challenges, particularly in dynamic environments like Fog-to-Things architectures. Threat actors may exploit these weaknesses to disrupt communication systems and undermine their integrity. Tackling these issues necessitates an intrusion detection system (IDS) that achieves a balance between accuracy, resource optimization, compatibility, and affordability. This study introduces an innovative deep learning-driven IDS tailored for fog-assisted smart farming environments. The proposed system utilizes a class-aware autoencoder for detecting anomalies and performing initial binary classification, with a SoftMax layer subsequently employed for multi-class attack categorization. The model effectively identifies various threats, such as distributed denial of service (DDoS), ransomware, and password attacks, while enhancing security performance in environments with limited resources. By utilizing the Fog-to-Things architecture, the proposed IDS guarantees reliable and low-latency performance under extreme environmental conditions. Experimental results on the TON_IoT dataset reveal excellent performance, surpassing 98% accuracy in both binary and multi-class classification tasks. The proposed model outperforms conventional models (convolutional neural network (CNN), recurrent neural network (RNN), deep neural network (DNN), and gated recurrent unit (GRU)), highlighting its superior accuracy and effectiveness in securing smart farming networks
Applying fuzzy Tsukamoto method to improve production efficiency in manufacturing industry
Get PDFManufacturing can increase competitiveness and reduce costs by improving production efficiency. The study’s goal is to develop a production prediction system using the fuzzy Tsukamoto technique. This method is used to model the uncertainty that occurs during the production process. Thus, production planning based on demand and inventory availability can be more accurate. After being tested on production data from a manufacturing company, the fuzzy Tsukamoto method showed the ability to make more efficient decisions than conventional methods. This system not only significantly reduces production costs but also improves overall operational efficiency, including resource management, waste reduction, and cycle time optimization. The adoption of this method provides added value to companies in facing increasing market competition while keeping production costs low without compromising quality