UDC 004.622
DEVELOPMENT OF HYBRID FACTOR ANALYSIS METHOD FOR SMALL MEDICAL DATA SET
O. A. Popova, techer, TGMU, Tjumen, Rossia;
orcid.org/0009-0006-3530-5703, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The paper presents a comparative analysis of the effectiveness of well-known methods of factor analysis: PLS, FastICA, BFA and MLFA, as well as newly developed hybrid methods PLS-NN and PLS-RF. The main aim of the study was to identify methods that provide the best accuracy in explaining the variance in target variable and the best fit of the model to data. The results showed that PLS method explained a significant proportion of the variance in target variable, demonstrating good model fit, but there were signs of excessive model complexity. FastICA method demonstrated high explanatory power, but potential overfitting problems were identified. BFA and MLFA methods showed unsatisfactory results, characterized by negative predictive performance and unsatisfactory values of model fit indicators. Based on the results of the study, PLS method was chosen for further improvement and adjustment. In order to increase its efficiency, hybridization was used, which significantly improved the quality of the model and its fit to the data. The analysis of the results of hybrid factor analysis methods (PLS-NN and PLS-RF) showed that both methods have high ability to ex plain variation in source data. However, PLS-NN method outperformed PLS-RF method in a number of indi cators, such as the coefficient of determination, information criteria AIC and BIC, as well as RMSEA and SRMR indicators, which indicates a better fit of the model to data and a lower level of approximation error. In summary, the study confirms that PLS-NN is the method of choice to be used in considered dataset due to its accuracy, explanatory power and model fit quality. The problem of finding the characteristics of data channel of aircraft opto-electronic trajectory measurements is studied. The aim is to find main channel qual ity indicators in a stationary mode with the transmission of priority and non-priority data opto-electronic means of trajectory measurements at different values of input stream intensity, window length and probabil ity of frame distortion during transmission. Input stream from each measuring station is simple with a given priority level. Frame transmission through the channel is based on window control. Channel quality indica tors using queuing system M/G/1 are found. The degree of frame error probability influence in case of transmission protocol window length while transmitting via communication channel on mediate values of frame number in a system, the time of frame waiting in a queue, determined for priority and non-priority frames is evaluated.
Key words: : medical data, factor analysis methods, PLS, FastICA, BFA, MLFA, hybrid methods of factor analysis, Random Forest, Neural Networks, coefficient of determination, information criteria, preprocessing of input data.