Optimisation of intelligent system algorithms for poorly structured data analysis

Integration of heterogeneous types of medical data using modern deep learning methods can improve the accuracy and efficiency of diagnosing complex diseases, such as cardiovascular diseases, which is relevant for personalised medicine and reducing the risk of medical errors. The study aimed to present the development of a decision support system for improving the diagnosis of cardiovascular diseases by integrating heterogeneous types of medical data. To create the knowledge base, data from real clinical scenarios were used, which underwent the stages of cleaning, standardisation, and semantic analysis using specialised medical dictionaries. The system demonstrated high efficiency due to its ability to integrate text, image and signal data into a single analysis process. The efficiency was evaluated by such metrics as accuracy, completeness, F1-score, and predictive values of positive and negative results. The introduction of transformers ensured a 15% increase in diagnostic accuracy compared to traditional methods, and the use of a hybrid computing approach reduced model training time by 30% and enabled the processing of up to 1 TB of data per day. Additionally, the integration of heterogeneous types of medical data into the system has improved the personalisation of diagnostics, accounting for individual patient characteristics such as medical history, genetic factors, or comorbidities. Transformer attention mechanisms improved resistance to noise and data gaps, which ensures reliable results even with incomplete or inaccurate information. Optimisation of the models reduced delays in data processing, which is critical for prompt clinical decision-making. In addition, transformers have proven their ability to dynamically scale to process new types of data without losing efficiency, opening opportunities