Search

Type 2 diabetes mellitus (T2DM) is one of the most common diseases and a leading cause of death. The problem of early diagnosis of T2DM is challenging and necessary to prevent serious complications. This study proposes a novel neural network architecture for early T2DM prediction using multi-headed self-attention and dense layers to extract features from historic diagnoses, patient vitals, and demographics. The proposed technique is called the Self-Attention for Comorbid Disease Net (SACDNet), achieving an accuracy of 89.3% and an F1-Score of 89.1%, having a 1.6% increased accuracy and 1.3% increased f1-score compared to the baseline techniques. Monte Carlo (MC) Dropout is applied to the SACDNet to get a bayesian approximation. A T2DM prediction framework based on the MC Dropout SACDNet is proposed to quantize the uncertainty associated with the predictions. A T2DM prediction dataset is also built as part of this study which is based on real-world routine Electronic Health Record (EHR) data comprising 4,124 diabetic and 181,767 non-diabetic examples, collected from 295 different EHR systems running in different parts of the United States of America. This dataset is further used to evaluate 7 different machine learning and 3 deep learning-based models. Finally, a detailed analysis of the fairness of every technique against different patient demographic groups is performed to validate the unbiased generalization of the techniques and the diversity of the data., Comment: Misspelled SACEDNet changed to SACDNet in Abstract. Related Work corrected tehcniques to techniques. Dataset rh replaced with Rh. Methodology, mod replaced with models, andrepresentation with representations. Removed bold formatting from Table 3 in Methodology. Results replaced Hence the with Hence, this

Sentiment analysis, or opinion mining, is a computational process to determine the polarity of a topic, opinion, emotion, or attitude. Most of the work done onsentiment analysis is for resource-rich languages, such as English and Chinese. However, only limited work has been done for Roman Urdu/Hindi, which is hence a resource-poor language. Developing a robust Sentiment analysis system for Roman Urdu/Hindi is necessitated due to two major reasons. First, Urdu/Hindi is the third largest spoken language in the world, with over 500 million speakers. Second, it is becoming increasingly used because people prefer to communicate on the web using Latin Script (26 English Alphabets), instead of typing using their language-specific keyboards.Since the work on Roman Urdu/Hindi sentiment analysis is still in its infancy stage, therefore an urgent development of new techniques and improvements inexisting techniques is required. In particular, the development of an automated technique to address the problem of Roman Urdu/Hindi text normalization is necessary as that widely affects the performance of all Natural Language Processing applications, including Sentiment classification. The non-availability of an annotated dataset is another major issue towards building effective techniques for Roman Urdu/Hindi sentiment analysis.In this thesis, challenging issues hindering the development of effective Roman Urdu/Hindi sentiment classification have been addressed. First, the largest-everdataset of 11000 Roman Urdu/Hindi reviews has been gathered from six different domains, using comprehensive annotation guidelines. Second, a machine learning-based Roman Urdu sentiment analysis is developed using different content-based features. Third, a novel feature selection technique, called Discriminative Feature Spamming Technique, has been developed for Roman Urdu/Hindi sentiment analysis. This technique identifies distinctive features based on a term utility criteria and then further increases