ДЕРЕКТЕРДІҢ НӘТИЖЕ БАҒДАРЛАНҒАН ТЕГІСТЕУ МОДЕЛЬДЕРІН ӘЗІРЛЕУ ЖӘНЕ ОЛАРДЫҢ ТИІМДІЛІГІН ТАЛДАУ

Abstract

This study examined the widely used Simple Exponential Smoothing, Moving Average, Random Walk, and Savitzky–Golay algorithms for smoothing fluctuating data. The effectiveness of the methods was assessed using metrics such as root mean square error (RMSE), mean absolute error (MAE), mean percentage error (MAPE), and signal-to-noise ratio (SNR).

The testing used Kazakhstan's gross domestic product (GDP) data for the period 1990–2024. These data were obtained from official World Bank sources.

The software module developed during the study allows users to smooth time series and discrete measurements from various sources using four different methods. The module allows users to experiment and determine the most effective parameters by selecting and changing the estimation parameters. In addition, the ability to visualize the smoothed data and export them to .csv, .xlsx, .json, .pdf formats is provided.

The main goal of the study is to combine commonly used smoothing algorithms and present them as an internal module of an intelligent information system. According to the results obtained, it was found that the Savitzky–Golay filter showed the lowest error (RMSE = 5.27) and the highest signal-to-noise ratio (SNR = 28.49 dB). The proposed smoothing module allows for effective use in preprocessing of time series and economic forecasting systems. The scientific novelty of the study is the combination of four data smoothing methods into one module and the provision of a unified architecture that allows users to control their parameters.