The forecasting of financial bubbles has been recognised as an important research topic in the applied finance literature. The objective of this study is to identify and predict financial bubbles in the BIST 100, BIST Bank and BIST Industrials indices over the period 2002M01-2024M06 by combining econometric and machine learning methods. Accordingly, the Generalised Supremum Augmented Dickey-Fuller (GSADF) test is first used to identify bubble periods. This test provides a more robust method for detecting bubbles by taking into account structural breaks and volatility changes in time series. Subsequently, the Random Forest Algorithm, a powerful machine learning method, is used to establish a predictive model for bubble occurrences by establishing correlations with a set of macroeconomic indicators. Moreover, the Random Forest algorithm determines the relative importance of financial and macroeconomic factors that influence the probability of bubble formation. Empirical findings from the analysis highlight that broad money, exchange rates, CPI, FED interest rate and Composite Leading Indicators Index are determinants of price bubble formation in BIST 100, BIST Bank and BIST Industrials indices in Turkey. Accordingly, it shows the importance of effective use of monetary policy and macroprudential measures to mitigate bubble risks and ensure financial stability. It also shows that machine learning methods have significant potential for early detection of financial bubbles and management of market risks. The evaluation of model performance reveals a high degree of predictive accuracy, reinforcing the efficacy of the proposed hybrid approach as a viable alternative in the realm of financial bubble forecasting.