Artificial neural network for predicting cashew nut (

Cashew production is affected by environmental dynamics and climate change. This study aimed to predict cashew productivity using an Artificial Neural Network (ANN) based on climate variables. The dataset comprised cashew productivity records from Selogiri Sub-district, Wonogiri, Central Java (1990–2020), along with annual rainfall, air temperature, and humidity over the same period. The analytical procedures included data collection, normalization, reprocessing, determination of network architecture using Mean Squared Error (MSE), parameter setting, training, denormalization, testing, and K-Fold cross-validation with Mean Absolute Percentage Error (MAPE). The optimal network architecture was 3:14:1, consisting of three input neurons (humidity, temperature, rainfall), 14 neurons in one hidden layer, and one output neuron (cashew productivity). This architecture yielded the lowest MSE of 0.0005. Validation using K-Fold produced an average MAPE of 40% with an accuracy of 60%. According to the MAPE criteria (20–50%), the forecasting model is categorized as fair or acceptable for predicting cashew productivity. These findings demonstrate the potential of ANN to support agricultural yield prediction under climate variability.