Predicting trends in a stock market is one of the most difficult and important tasks in finance, as it directly influences investment strategies and outcomes. Recently, machine learning techniques have been applied to model the time series of the stock prices, which are more accurate and adaptable in a certain sense than traditional methods. In this review paper, we take a look at several ML approaches for prediction in the stock market, specifically tree-based models, deep learning architectures like long short-term memory (LSTM), and hybrid approaches that use external information, such as sentiment analysis. We analyse recent developments, pointing out strengths, weak points, and areas of gaps that could be of interest for future research. Our findings indicate that although deep learning models are promising, particularly hybrid frameworks, significant problems in the context of overfitting, non-stationary data, and real-time predictions remain.