This necessity has been amplified by the emergence and advancement of Artificial Intelligence (AI) algorithms – a technology demonstrating unparalleled potential in modeling and predicting complex economic and financial phenomena. In response to this critical need, the present research introduces a comprehensive framework for financial risk prediction and portfolio optimization. The core approach of this study involves a strategic integration of powerful Machine Learning (ML) algorithms, advanced Long Short-Term Memory (LSTM) neural networks renowned for time-series data analysis, and the K-Means clustering technique for uncovering hidden patterns within the data. The research dataset comprises detailed stock return data from companies operating on the Tehran Stock Exchange over a five-year period, from April 2020 (Farvardin 1399) to March 2025 (Esfand 1403). The initial phase of analysis was dedicated to clustering the return data using the K-Means algorithm. This process resulted in the identification of three distinct and optimal clusters, indicative of grouped stock return behaviors under various market conditions. In the subsequent step, the predictive performance of three prominent algorithms – LSTM, XGBoost (as a robust tree-based model), and multiple regression (as a classic statistical method) – was rigorously evaluated. The findings from this assessment revealed the significant superiority of the LSTM algorithm; achieving an accuracy of 91%, this model not only excelled in predicting stock returns but also registered the lowest error rate compared to the other models. Finally, to assess the risk management aspect, the classic Markowitz portfolio optimization model was implemented using the outputs from the LSTM algorithm. This implementation demonstrated that a portfolio constructed based on LSTM predictions experienced the lowest level of risk. Overall, this study pursues two primary objectives: first, a comparative evaluation of the predictive capabilities of Machine Learning and Deep Learning algorithms; and second, an assessment of the overall impact of employing AI tools in calculating and managing portfolio risk compared to traditional methodologies. The findings of this research strongly emphasize the advantages of utilizing AI algorithms in achieving more optimal and efficient outcomes in the domain of financial risk management.