This research presents a novel hybrid approach that amalgamates the strengths of Gradient Boosting Decision Tree (GBDT) and Artificial Bee Colony optimization (ABC) techniques for accurate stock price prediction. The features of the real-world data are of higher dimension and greatly redundant. Identification of informative features have become a vital step in data processing. The novel feature selection technique increases the efficiency and information of the features selected. We utilize the bee colony optimization technique to recognize informative features, hence attaining the best features of decision tree inputs in our approach. The integration of GBDT's ensemble learning capability with ABC's feature selection process aims to increase the predictive accuracy and fitness of stock price forecasts. The main objective of this paper is to provide a hybrid model that addresses the challenges of volatility, nonlinearity, and noise in financial markets, providing a potential avenue for improved decision-making in investment strategies. Features with lesser significance are suppressed. Experiments are conducted using three different markets for verification of the proposed model. The suggested approach is noted to demonstrate better result.