This research presents a deep learning approach for predicting durian prices using Bidirectional Long Short-Term Memory (BiLSTM) networks with seasonal analysis. The study addresses two critical challenges in agricultural price forecasting which include handling missing values in seasonal time series data and capturing complex temporal dependencies. The proposed methodology employs Prophet model for missing value imputation incorporating multiple seasonal components and Fourier terms to maintain temporal integrity. A BiLSTM architecture is implemented with a 25-day sliding window optimized to capture market cycles and seasonal patterns. Therefore, the model is evaluated using 10-fold cross-validation on durian price data from Thailand’s primary trading hub during 2023-2024 harvest seasons. The findings demonstrate strong predictive performance with R-squared values of 0.9254 ± 0.0439 on validation data and Mean Squared Error of 65.3470 ± 33.1819. The model successfully captures key market phases including initial stability, early decline, mid-season plateau and peak season dynamics. The integration of seasonal features with bidirectional processing enables accurate prediction of price movements providing valuable insights for stakeholders across the durian supply chain.