The increasing complexity and volatility of financial markets necessitate the use of advanced machine learning techniques for making accurate, real-time predictions that can drive informed trading decisions. This study applies Light Gradient Boosting Machine (LightGBM), a powerful gradient boosting algorithm, to market action prediction, combined with K-means clustering to uncover hidden patterns in market data. The primary problem is predicting whether a market action (such as buy or no-buy) should be taken based on historical market features such as price changes, volatility, and momentum. After training and validation on a dataset of historical market indicators, the LightGBM model achieved a cross-validation accuracy of 90.81%, with balanced precision, recall, and F1-scores across both classes (“Action” and “No Action”). The area under the receiver operating characteristic curve was 0.97, indicating that the model strongly distinguishes between classes. In addition to supervised learning, the data were grouped into three clusters using K-means clustering. Although the clusters were balanced across the “Action” and “No-action” labels, further sub-clustering within individual clusters revealed more nuanced patterns that could aid the refinement of trading strategies. The results suggest that LightGBM is an effective tool for real-time market prediction, while clustering analysis offers additional layers of insight into market behaviors. This study has important implications for algorithmic trading, risk management, and decision-making in financial markets, highlighting that machine learning can potentially enhance predictive accuracy and uncover new strategies in market analysis.