This article contains methods, results, and analysis agreed for the development of an application based on the internet of things and making use of machine learning techniques that serves as a support for the identification of the saline wedge in the Magdalena River estuary, Colombia. As a result of this investigation, the process of identifying the most suitable telecommunications architecture to be installed in the estuary is shown, as well as the characteristics of the software developed called SISME (Estuary Monitoring System), and the results obtained after the implementation of prediction techniques based on time series. This implementation supports the maritime security of the port of Barranquilla since it can support decision-making related to the estuary. This research is the result of the project “Implementation of a Wireless System of Temperature, Conductivity and Pressure Sensors to support the identification of the saline wedge and its impact on the maritime safety of the Magdalena River estuary”.

This paper addresses the limitations of existing deep learning techniques in retinal disease diagnosis from fundus images, specifically focusing on inadequate feature representation, localization challenges, and the lack of interpretability. Current methods also struggle with class imbalance and the need for simultaneous multi-task learning. To overcome these issues, we propose a novel interpretable deep learning framework for the simultaneous segmentation of multiple retinal lesions and multi-categorical disease classification with grading. Our framework incorporates a modified Convolutional Neural Network (CNN) with separable convolution layers for efficient parameter utilization and an attention mechanism to focus on critical image regions, enabling accurate lesion mapping. To tackle class imbalance, the framework utilizes the Synthetic Minority Over-Sampling Technique (SMOTE). Furthermore, Explainable AI (XAI) techniques, including Grad-Cam, are integrated to enhance the model's interpretability and provide insights into its decision-making process. The effectiveness of the proposed framework is evaluated on several publicly available datasets, including ODIR, RFMiD, and IDRiD. The framework aims to provide a comprehensive, balanced, and interpretable solution for real-time fundus disease diagnosis in healthcare. The proposed architecture outperforms existing models in accuracy, AUC, mAUPR, and mIoU on retinal fundus image datasets. Heatmaps and segmentation masks confirm their effectiveness in identifying disease regions and segmenting lesions.

Social media sites have developed into a common space for individuals to share their concerns and opinions. There is a chance for individuals and organizations to participate in online behavior that breaches accepted social norms because of the preservation of anonymity and the freedom to communicate ideas without restriction. This leads to a rise in the degree and intensity of hate speech in the online environment. Urdu is the national language of Pakistan and is also widely spoken across several other countries, with over 170 million speakers worldwide. This research addresses the detection of hate speech in Roman Urdu, a prevalent language in Asia, where limited resources exist for mitigating hate speech compared to English. Leveraging machine learning, deep learning, ensemble learning, and natural language processing, we developed a system proficient in understanding Roman Urdu language and culture, capable of identifying diverse hate speech manifestations like abusive language, religious hate, sexism, and racism. We expanded the Roman Urdu Hate Speech and Offensive Language Detection dataset to encompass 30,955 instances, incorporating a novel “Racism” category. Our dataset includes various classes of hate speech such as abusive/offensive, religious hate, sexism, and racism, each reflecting distinct patterns of discriminatory language prevalent in Roman Urdu. After executing text pre-processing, we utilized feature extraction techniques such as Bag of Words and Term Frequency-Inverse Document Frequency embeddings. For model building, we employed several supervised machine learning algorithms, including Random Forest, Decision Tree, Multinomial Naive Bayes, Support Vector Machine, and ensemble methods, coupled with K-Fold cross-validation for robust validation. Additionally, unsupervised learning techniques such as the Gaussian Mixture Model and k-means clustering were also implemented. Deep learning approaches, including Bidirectional Encoder Representations from Transformers, Convolutional Neural Networks, Long Short-Term Memory networks, and multilingual BERT, were explored. Among these, mBERT distinguished itself by achieving an impressive accuracy of 92%, notably surpassing the baseline performance.