This paper explores recurrent neural network architectures that integrate fuzzy logic principles for financial time series forecasting. Financial time series forecasting is particularly challenging due to high volatility and nonlinear dependencies. To address this, we employ a Long Short-Term Memory (LSTM) model, widely recognized for its ability to handle sequential data. Fuzzy logic is incorporated both during data preprocessing and within the model, providing a flexible representation of input features, reducing noise sensitivity, and emphasizing relevant patterns. This adaptive mechanism tailors the model to the characteristics of specific financial data, capturing complex dependencies while mitigating noise. Experimental results demonstrate that the proposed approach outperforms the standard LSTM model in predictive accuracy.