This study presents a comparative analysis of the hybrid model Mamba (Memory-Augmented Model for Biased Attention) and Transformer model for predicting the population dynamics of pests of the bread striped flea (Phyllotreta vittula) in Kazakhstan. The objective is to evaluate their effectiveness in agricultural forecasting by assessing key performance metrics, including Mean Squared Error (MSE) and R2 (Coefficient of Determination). The hybrid Mamba model, which integrates Selective State Space Layers (SSBLayer), demonstrates superior accuracy in forecasting pest population dynamics, particularly in scenarios with limited training data. Experimental results reveal that the Mamba model consistently outperforms the Transformer model, achieving lower MSE and higher R2, making it a more effective tool for pest prediction. The findings suggest that incorporating environmental variables, such as temperature, humidity, and soil conditions, could further enhance prediction accuracy. This research contributes to the advancement of machine learning applications in agriculture, facilitating timely pest detection and improved crop protection strategies. Future studies could explore the potential of the proposed Mamba-based model for predicting other pest species and its applicability across different agricultural regions, ensuring broader generalizability.