This paper proposes a novel method to stock market forecasting based on long short-term memory (LSTM) models, which are recurrent neural networks designed to record and evaluate temporal correlations in sequential data. LSTMs are ideal for detecting detailed patterns in previous market data, allowing them to accurately forecast future stock prices. To improve the accuracy of these forecasts, the model includes exogenous inputs such sentiment analysis of financial news headlines, company reports, and social media trends. The model also takes a multitimeframe approach, assessing market movements across short, medium, and long periods. This multidimensional study aids in the identification of trends that are only observable during certain time periods, resulting in a more complete knowledge of market behavior. The suggested strategy dramatically outperforms standard market prediction algorithms, which sometimes depend entirely on historical market data or linear statistical methods. The findings demonstrate how machine learning approaches, particularly LSTM models, may greatly enhance the accuracy and reliability of stock market projections, providing vital information for investors, financial analysts, and policymakers.