FEATURE ENGINEERING FOR THE IMPLEMENTATION OF MACHINE LEARNING IN CLINICAL DATA PROCESSING

This paper presents a study of feature engineering for the application of machine learning (ML) in clinical data processing, focusing on binary classification of time series data. The study demonstrates the effectiveness of using the Haar transform to enhance feature importance and improve classification performance. The Haar transform allows for increased predictive accuracy by augmenting the weight of significant features, which is especially crucial in handling complex clinical data. The research results show a substantial increase in the area under the receiver operating characteristic curve (AUC-ROC) from 0.44 for the baseline model to 0.82 for the Haar transform model, indicating significant improvements in predictive accuracy. The methodology described in the paper encompasses various stages, including data preprocessing, model training using the XGBoost algorithm, and performance evaluation via AUC-ROC curves. Data preprocessing involves cleaning and normalizing the data, critical steps to ensure high-quality machine learning outcomes. Special attention is given to using Internet of Things (IoT) data in clinical settings, which opens new possibilities for predictive analytics and decision-making in healthcare. The approaches described in the paper can be utilized to analyze large amounts of information collected from various medical devices connected to the IoT network. This allows for more accurate predictions and informed decisions based on real data, contributing improving of medical services and patient care quality. The research results underscore the potential of machine learning methods in healthcare institutions to enhance predictive accuracy and decision-making. Future research directions may include exploring additional feature engineering methods and using advanced machine learning algorithms to further increase the utility of clinical IoT data analytics. In particular, exploring the possibilities of deep learning and neural networks may open new horizons for clinical data analysis and processing.