Search

There are growing numbers of significant skin disorders, including skin pigmentation. It states that skin color is determined by the amount of melanin produced by the body. The two main categories of skin pigmentation are hyperpigmentation, in which pigment seems to overflow, and hypopigmentation, in which pigment appears to decrease. However, many skin conditions share characteristics, making it difficult for dermatologists to correctly early diagnose their patients. Consequently, the accurate detection of skin disorders and the diagnosis of dermatoscopy pictures can be greatly aided by machine learning and deep learning approaches. The most effective deep learning technique for picture identification was investigated to diagnose hyperpigmented skin diseases. YOLO, DenseNet201, GoogLeNet, InceptionResNetV2, and MobileNet were among the pretrained models used to classify four most common hyperpigmented skin disorders. Using assessment metrics like accuracy and AUC (Area Under Curve), it was determined which deep learning method would work best for creating a clinical diagnostic system. The study analyzed the accuracy rates of five pretrained models, including GoogleNet, MobileNet, DenseNet201, InceptionResNetV2, and YOLO, after 50 iterations. Using metrics such as accuracy and Area Under the Curve (AUC), the study evaluated the models' performance on a small dataset split into 80% for training and 20% for testing. Training accuracy rates were 93.8%, 100%, 100%, 98.77%, and 97.43%, respectively, while test accuracy rates were 87.18%, 79.49%, 87.18%, 89.74%, and 97.56%. DenseNet201 showed strong performance, particularly for cafe-au-lait spots (CS), melasma (ML), and nevi (MN), but struggled with congenital nevus (CN). Despite DenseNet201′s strong traditional CNN capabilities and generalization, YOLO emerged as the top model due to its stable accuracy and AUC values, as confirmed by confusion matrices. While these models show promise as diagnostic tools for dermatologists, further research, including expanding the dataset and exploring hybrid models, is needed to enhance their clinical accuracy and effectiveness. [ABSTRACT FROM AUTHOR]

Non-player Characters (NPCs) play a crucial role in shaping the player's experience and immersion in video games. However, despite advancements in technology and game design, NPCs continue to face several challenges that hinder their ability to provide engaging and realistic interactions. This has been true in some cases, such as open world games where the random NPCs do not have a lot of character depth or use repetitive sentences. This research aims to improve dialogue in video games by humanizing conversations between players and NPCs by using Google's Gemini API and Sentence-BERT framework. Gemini 1.0 Pro as a base model from Google to give natural responses and Sentence-BERT framework based on PyTorch and Transformers to compute sentence or text embeddings for more than 100 languages that can be compared with cosine similarity algorithm to find sentences with similar meanings. NPCs will be able to provide creative and unique quests to give to the players based on their conversations. This research provides insight on how LLMs such as Google's Gemini and Sentence BERT can be used in role-playing games to generate dynamic, unique conversations for NPCs without having to make the script manually [ABSTRACT FROM AUTHOR]

Brain Tumor is the abnormal growth of cancerous cells in the brain. Being able to diagnose them quickly allows practitioners to take preventive measures early, increasing survival rate. Currently, most research utilizes the popular CNN for classification. In this study we aim to find a better classifier by implementing an ensemble method, specifically using CNN and XG-Boost. The dataset we used consists of 3652 MRI images under 2 classes: tumor and no tumor. Our experiment starts off by developing a CNN model. We then used the model for feature extraction and trained an XG-Boost Classifier on the extracted features. The results show an improvement from a 97.61% to a 99.43% testing accuracy going from the CNN to the ensemble model. This demonstrates the potential of ensemble methods for enhancing standard CNN models for tasks like this. This could also open avenues for further advancements in clinical image diagnosis by utilizing the potential of deep learning in combination with ensemble learning. [ABSTRACT FROM AUTHOR]

The rising prevalence of digital transactions in contemporary financial systems has heightened the risk of credit card fraud (CCF), necessitating effective detection techniques. Machine learning (ML) technologies show promise, but dataset imbalance impedes fraud detection accuracy. This study investigates the effectiveness of ensemble learning techniques and data balancing methods for detecting CCF in imbalanced datasets. We compare four ML algorithms—Logistic Regression, Decision Tree, Random Forest, and XGBoost—and evaluate data balancing techniques including SMOTE, ADASYN, Random Oversampling, CNN, NCR, and Random Undersampling, along with hybrid methods like SMOTE with NCR and ADASYN with NCR. Our experiments use a dataset of credit card transactions made by European cardholders, which includes 30 attributes and exhibits severe class imbalance, with fraud events accounting for only 0.172% of total transactions. Applying balancing techniques led to significant improvements: Random Oversampling with XGBoost achieved the highest F1-Score of 92.43%, with Precision of 97.383% and Recall of 88.418%. Random Forest with Random Oversampling followed with an F1-Score of 91.554%, and NCR with Random Forest achieved an F1-Score of 91.508%. These findings highlight the effectiveness of oversampling techniques in enhancing model performance for imbalanced datasets. Precision, Recall, and F1-Score were emphasized, while specificity and sensitivity were not directly used; macro averages ensured balanced evaluation across all classes. This study concludes that combining Random Forest or XGBoost classifiers with oversampling techniques like Random Oversampling or undersampling methods such as NCR provides accurate and unbiased results for detecting CCF in imbalanced datasets, contributing to reliable fraud detection systems for real-world applications. [ABSTRACT FROM AUTHOR]

This study aims to find out the impact of data augmentation and Synthetic Minority Over-sampling Techniques (SMOTE) implementation on the classification models performance using small and imbalanced dataset of student academic performance. The design of this study involved a comprehensive experiment by comparing four scenarios: 1) comparing classification models without both data augmentation and SMOTE, 2) models with data augmentation, 3) models with SMOTE, and 4) models with both data augmentation and SMOTE. The model's performances were each measured based on standard evaluation metrices such as accuracy, precision, recall, F1-score. To test the results validity, there were three classification algorithms implemented and evaluated for each scenario, that is, Random Forest, XGBoost, and AdaBoost. The finding of this study highlights the significant impact of data augmentation and SMOTE to the increase of classification models performance, particularly over the small and imbalanced dataset. Results showed that the implementation of both techniques simultaneously brought about the most significant increase in the evaluation metrices compared to the implementation of both techniques separately. The originality of this study lies in its comprehensive approach in comparing the effectiveness of data augmentation and SMOTE, as well as the use of student academic performance dataset, which is, a real case in the context of artificial intelligence. This finding gives a valuable insight to the researchers and practitioners in choosing appropriate techniques to handle small and imbalanced class datasets. This study is expected to make an important contribution to the more effective development of classification methodology in various domains. [ABSTRACT FROM AUTHOR]

The aim is to develop a machine-learning model that uses uneven data to predict child nutrition. Methods: We utilized an Electronic Medical Record (EMR) that included weight, height, sex, and age data related to toddlers. We used Random Forest (RF), XGBoost, and Support Vector Machine (SVM) in our analysis. Imbalanced data was handled using the Synthetic Minority Oversampling Technique (SMOTE). Results: SMOTE-balanced data with RF performed better on almost every performance metric, including f1-score, accuracy, precision, recall, and ROCAUC, by 84.70%, 88.90%, 80.00%, 84.20%, and 93.00%, in that order. In conclusion, SMOTE improved the classifier system's ROC AUC, f1-score, accuracy, precision, recall, and recall, according to our research. [ABSTRACT FROM AUTHOR]