Аннотация
When making stock market decisions you need to analyze a ton of news to figure out what the overall sentiment towards stocks is. While there are good sentiment analysis models out there, none are for Kazakhstan specific financials. To fill this gap we have gathered a custom dataset of Kazakhstan focused financial news and trained machine learning models to test them. This paper explores the impact of financial news headlines on Kazakhstan stock markets using machine learning models. We classify news sentiment into three categories (-1: negative, 0: neutral, 1: positive) and compare traditional machine learning models with transformer based models. Our results show that logistic regression gets 70% accuracy while transformer models outperform classical approaches with 75-80% accuracy. This shows the power of deep learning in capturing subtle financial sentiment making transformer based models more suitable for real world financial applications even with higher computational cost.