Hybrid heuristic-based optimal weighted fused feature for convolutional long short-term memory-based intelligent crop yield prediction model.

Agriculture is a highly important sector of the Indian economy with half of the individuals relied on the subsistence of food crop production. Since agriculture is an emerging field in the economic world, crop identification, crop prediction, and classification, it reached huge attention. Particularly, crop yield prediction attracts scholars to develop a model of it to help the country in an economical way. Due to the vulnerability of the environment and soil conditions including soil quality, humidity, rainfall, and temperature as well. Farmers are in the position to make the decision of crop cultivation, monitoring its growth and harvesting time. Yet, the fluctuation of the climatic state makes it critical for farmers to predict crop yields. Also, the less moisture in the soil misleads the prediction results. These all come to the end of determining the relationship between food security and import–export strategy. Hence, the development of novel prediction systems is necessitated in agriculture industries. To conquer such shortcomings, a hybridized deep learning model is developed for crop yield prediction. Firstly, the required raw data is collected from benchmark datasets and utilized for subsequent sections. Secondly, the noteworthy features are extracted by using the autoencoder, statistical feature, and Principle Component Analysis (PCA). Thirdly, these resultant features are offered to acquire the optimal weighted features, where weight is optimized by proposing the novel algorithm as Fitness-derived Adaptive Harris Hawks Optimization (FA-HHO). Finally, the obtained weighted features are subjected to Hybrid Deep Learning (HDL), which is constructed with the 1-Dimensional Convolutional Neural Network (1-DCNN) with a Long Short-Term Memory (LSTM) layer and Deep Temporal Convolutional Neural Network (DTCNN). Further attaining the optimal values, the hyperparameters are tuned optimally by the FA-HHO approach. The final prediction rate is estimated by taking the average value. The performance of the prediction model is evaluated using diversification of parameters and compared against baseline approaches. Hence, the findings reveal that it contains the potential to predict the crop yield to help the farmers to increase productivity. [ABSTRACT FROM AUTHOR]