In financial time series analysis, accurately detecting change points is crucial for effective forecasting and risk management. This study investigates the impact of different colored noises on the performance of change point detection algorithms, focusing on the CIR model with LSTM and Catboost classifiers. We evaluated the algorithms’ ability to forecast and classify change points versus normal points using a number of previous observations, comparing the effects of white, red, pink, blue, and violet noise. Additionally, this paper discusses numerical approximation using the implicit Milstein scheme, parameter estimation techniques, and the PELT algorithm used to find change points in a time series. These findings highlight the critical role that noise color plays in enhancing the effectiveness of change point detection in financial time series and underscore the importance of methodological considerations in achieving optimal performance.