8333 research outputs found
Sort by
Exploring Machine Learning Algorithms for Automated Segmentation of Brain Tumors from MRI Scans
No full textThe paper evaluates the potential of using pre-trained Convolutional Neural Network (CNN) models for the automatic segmentation of brain tumors from MRI scans. Traditional brain tumor segmentation techniques are time-consuming and highly variable, highlighting the need for faster and more accurate methods. This study explores the performance of three transfer-learning-based pre-trained CNN models: MobileNetV2, VGG16, and ResNet50. These models were trained and tested using augmentation techniques for increasing diversity on an MRI image dataset. ResNet50 achieved the highest test accuracy at 81.48%, followed by VGG16 at 77.78%, and then MobileNetV2 at 59.26%. The result of the current study proves that pre-trained models can feasibly be used for the segmentation of brain tumors but require further optimization in order to improve their accuracy and generalization. Given the diversity of the datasets and techniques in reducing overfitting, such models can potentially improve diagnoses in neuro-oncology. Commercialization by fully automated tools that seamlessly integrate into the routine clinical workflow is foreseeable, saving time for radiologists and improving patient outcomes
Exploring the Economic and Social Aspects of Youth Smoking: A Multi-Dimensional Analysis
No full textYouth using tobacco product is connected to long-term health risks, like cardiovascular disease, respiratory illnesses, and various cancers. The study revolves around investigating the influence of household income, healthcare utilization, financial pressures, education, and geographic factors on youth tobacco use. The data used for this research is from the Youth Tobacco Survey (YTS), Behavioural Risk Factor Surveillance System (BRFSS), and Annual Social and Economic Supplements (ASEC), in this study various advanced machine learning models were applied like Random Forest regressor and XGBoost (a gradient boosting algorithm). The findings showed that the socioeconomic factors, particularly family income and food security, significantly impact youth smoking behaviour. whereas, XGBoost outperformed other models in predictive accuracy, while giving robust insights into these complex interactions. The Geospatial analysis was done to identify regions with higher smoking rates and with help of that those finding targeted interventions can be done on those high-risk areas. These results show valuable insights and route for policymakers aiming to reduce youth tobacco use, underscoring the need for comprehensive, data-driven public health strategies. Also, as the socioeconomic is also a factor that’s influencing the youth smoking, this study will help to target the areas with high risk and with the geospatial analysis and by this multi-dimensional approach from machine learning models and data from different sources will help to understand the factors and this study will help in public health policies while, considering all the points that could change the dynamics. This is further evidence of the significant and ongoing impact that socioeconomic elements have in influencing tobacco use among young people. The study supports this claim by using sophisticated machine learning and geospatial analysis to create one integrated model that takes into account family income, food security level, accessibility of parks for physical activity time (PIC), along with other demographic information contribute best or worse to smoking. The knowledge achievable in this study can be a key to public health interventions and policies targeting the decrease of tobacco use among youth, taking special notice of those geographic areas found at risk through geospatial analysis. Together, this versatility provides a strong foundation to tackle this substantial public health concern
How Well Can MobileNetV3 Perform in Detecting Diseases in Tomato Plants
No full textIn this study, we used the PlantVillage dataset to assess the performance of the MobileNetV3 model in identifying diseases in tomato plants. Our foremost objective is to develop a model that can not only achieve high accuracy but also operates systematically, making it practical for deployment in agricultural sectors with limited hardware resources. We’re aiming to enhance the models diagnostic capabilities by employing transfer learning optimisation. We evaluated the performance of this model by using recall, accuracy, precision and F1 score. This research aims to highlight and prove the potential for integrating deep learning with agriculture technology to promote sustainable farming practices through better disease detection
A VMD and FAN Based Hybrid Model for Air Quality Index Forecasting
No full textThe time series data of Air Quality Index (AQI) is very complex and nonstationary, therefore the forecasting and accurate prediction of AQI is challenging. This study propose a novel VMD-FAN hybrid model using Variational Mode Decomposition (VMD) for handling noise and Fourier Analysis Networks (FAN) for handling periodicity, the hybrid model is good at predicting short term time series AQI data on a air quality dataset of Taiwan. The cleaned AQI timeseries extracted from the original dataset is decomposed into individual Intrinsic Mode Functions (IMFs) using VMD and each IMF is predicted using a FAN model subsequently aggregated to form a final forecast of AQI values. The proposed hybrid model predicts the AQI of Annan district in Taiwan with a MAE, MSE, RMSE and MAPE as 0.717643, 1.352704, 1.163058, 1.495354% respectively and is better than the compared base model. The generalizability of the model is further validated with extension analysis on different cities in Taiwan. The proposed hybrid model showcase high performance and its ability to predict complex AQI data and contributes to the research in the domain
Decoding Online Pharmacy Trends: Clustering, Prediction and Business Insights
No full textThe online pharmacy market in India is experiencing significant growth, driven by increasing consumer demand for convenience and accessibility in healthcare services. In this competitive market, it is essential to extract insights from available data, identify areas for improvement, and enhance business operations and services. This study aims to uncover patterns within online pharmacy data sourced from the Kaggle platform. Analysing biomedical text is challenging but domain-specific transformer-based models are effective in extracting relevant entities in such scenarios. To achieve this, named entity recognition (NER) models such as Med7 and Clinical-AI-Apollo (Medical-NER) are used to extract features. The data is analysed using K-Means clustering, and a classification model is built to predict product reviews using supervised machine learning models Random Forest, XGBoost, and Easy Ensemble classifiers. For vectorization, transformer-based models such as BioBERT, BioFormer-16L, and Clinical-AI-Apollo are used. Evaluation results show that the Easy Ensemble classifier with XGBoost estimator effectively handles class imbalance, and when combined with the Clinical-AI-Apollo model for vectorization, it outperforms other models in terms of ROC AUC performance. This work contributes to a deeper understanding of the data, providing business insights into identifying disease profiles, side-effects, medicine forms, composition and manufacturer etc. along with the review-based performance probability of products. These insights can inform better strategies to enhance underperforming products
Enhancing SMS Spam Detection using Deep Learning Techniques
No full textSMS spam is a major problem in mobile communication which will cause issues like financial loss and privacy violations. SMS spam detection is used to identify unwanted messages, protecting users from scams. The primary goal of SMS spam detection is to identify the difference between legitimate (ham) and unwanted messages(spam). Many models have been used to detect spam, but due to advancements in spam techniques we needed better detection methods to identify the advanced spam messages. Traditional machine learning models like Random Forest and deep learning models such as RNN, LSTM, BI-LSTM, and GRU have worked well in identifying spam messages, but it fails to understand the deeper meaning of the message. This research focuses on identifying these issues by using advanced Transformer models like Bert to compare whether these models will perform better than the traditional methods and other deep learning models. This research is very important because due to the rise in technology development spam messages are more sophisticated to identify, so we need powerful and accurate detection models to detect them. People use their mobile a lot for communication, transferring information and protecting them from unwanted messages will improve their security and user experience. Transformer models like Bert are good in understanding the deeper meaning of the word, so this research identify whether these models can expect to do a better performance compared to other methods in detecting SMS Spam
Wrong-Way Vehicle Detection Using YOLOv7 for Enhanced Traffic Safety
No full textWrong-way driving is a significant contributor to road accidents and traffic congestion worldwide. Traditional methods for detecting wrong-way vehicles, such as manual monitoring, fixed cameras, or traffic sensors, have limitations in terms of real-time detection, scalability, and accuracy. These traditional systems often fail to provide timely alerts, especially in dynamic traffic conditions. This study addresses these gaps by implementing an advanced vehicle detection system using YOLOv7, which can accurately identify wrong-way drivers in real-time. So, the dataset used for this study is the ”Vehicle Detection” dataset, which contains a diverse set of images representing different vehicle types, including ambulances, buses, cars, motorcycles, trucks, and vans. The primary objective of this study is to develop a robust vehicle detection system capable of identifying and tracking vehicles in real-time video streams. Several models were explored for this task, including YOLOv5, YOLOv8, and YOLOv7, with each model trained and tested on the vehicle dataset. Among these, the best performance was achieved using YOLOv7, which demonstrated the highest [email protected] score i.e. 0.876, making it the optimal model for this vehicle detection task. YOLOv7 outperformed other models in terms of accuracy and precision, particularly excelling in detecting various vehicle classes, such as cars and trucks, with higher precision and recall values. The study also included the implementation of a real-time detection system that tracks vehicles and identifies wrong-way driving violations using a reference direction vector and the system is evaluated based on assumptions using sample traffic videos
Tailoring Customer Engagement: Advanced Segmentation for Growth in Garden Business
No full textThe garden industry is comprising of the seasonal sales patterns and eco-conscious customers, faces challenges to segment the customers who are diverse in their characteristics. This research explores the advanced data mining techniques like K-means, DBSCAN, Hierarchical Clustering, and GMM algorithms integrated with RFM analysis to address the challenges. Using the customer purchasing details of the company “My Dream Garden”, the study focuses to evaluate the performance of the algorithms using Silhouette Score, Elbow Method, and Davies-Bouldin Index. The research described DBSCAN method as the most effective to segment the garden customer showcasing the highest Silhouette Score (0.588) confirming that the algorithm can handle noise and irregular cluster shapes. GMM exhibits a score of 0.502 which is a probability clustering while the Hierarchical clustering exhibits a score of 0.477. Despite the popularity of the K-means, this method underperformed for the garden industry to segment the customers with the lowest score of 0.278 struggling hard with the non-linear data. RFM analysis skillfully categorises the customers to “Best Customers” and “At Risk Customers” underlining the need to apply strategies on personalised marketing, customisation and to retain the customer. While DBSCAN proved robust but its reliance on manual parameter tuning and the study's focus on transactional data reveal opportunities for refinement. This research provides a robust framework for customer segmentation in the gardening sector, empowering businesses with insights to enhance customer satisfaction, loyalty, and revenue
Predicting Airline Passenger Satisfaction with Stacking Classifiers and Machine Learning Models
No full textIn the airline industry, the important aspect that impacts an airline’s performance is the satisfaction of passengers. Airlines that can understand and satisfy their passengers’ expectations succeed through customer loyalty and increased sales. In addition to ticket pricing, modern passengers assess the kind of services provided by various carriers before making their choices. Therefore, airlines need to ensure service excellence. Passenger feedback leads to these expectations, and consequently, the airlines will take note of these for fine-tuning. This research has performed data mining on a dataset of over 130,000 customer satisfaction ratings for various airlines to analyze the main drivers of satisfaction. Even with increasing competition, most studies in the past failed to address the complexity of passenger satisfaction, often limiting either the factors or traditional methods used. This study focused on bridging the gap that identifies and predicts satisfaction drivers by embedding stacking classifiers and machine-learning models. The best performance was obtained for the Stacking model with the meta-learner classifier MLP, with a 96.53% accuracy, 97.97% precision, 94.06% recall, 95.97% F1 score, and 92.99% MCC. The study has underpinned key fact-oriented decision-making in achieving high satisfaction of customers in the airline sector to gain a competitive advantage
Enhancing Real-Time Fire Detection with RT-DETR and Optimized Dataset Preparation
No full textReal-time fire detection systems face significant challenges achieving high accuracy and efficient processing speeds. This research looks at the optimization of dataset preparation techniques that could be used to implement RT-DETR in fire detection systems, together with augmentation strategies. The presented study aims to bridge the critical gap between dataset preparation methodologies and the performance of transformer-based architectures in safety-critical applications.
In the work, a holistic approach was followed with the RT-DETR-L architecture. Extensive data augmentation is done through geometrical transformations, including changes in intensity. The implementation is performed on Google Colab, running on an A100 40GB GPU infrastructure. It comes with a dataset of 2,200 validation images. The pipeline is carefully designed in such ways that the aspect ratios of the images are preserved, standardized at 640x640 resolution.
Very remarkable performance metrics are presented, with 0.985 for mAP@50 and 0.949 for mAP@50-95. The system also retains very high precision regarding fire detection at 0.991 and smoke detection at 0.962, with only 16.6ms of processing time per image. This represents substantial outperformance compared to the current benchmarks while preserving real-time processing capabilities.
These results provided new baselines for real-time fire detection systems and came with useful insights into the optimization of dataset preparation for transformer-based architecture. The contribution of this study both in theoretical understanding and practical implementation strategies advances the development of enhanced fire detection systems, which shall be very operational, especially in application areas concerning the safety of human life where a real-time response is expected