Predicting Crops based on Soil Features and its Surroundings Using the Capsule A-BiLSTM Method

Abstract

Traditionally, farmers grew crops based on their knowledge and experience. The weather has changed, though, and it's hurting food yields. Because of this, farmers can't pick the right crop(s) based on the soil and the climate, and trying to guess which crop(s) would grow best on their land by hand has usually failed. More crops are grown when crop predictions are accurate. Here, machine learning is very important for making predictions about crops. Predicting crops relies on the soil, the location, and the weather. Choosing the right attributes for the right crop(s) is an important part of how feature selection methods make predictions. Putting together all of the features from the raw data without first checking what part they play in the model-making process will make it more complicated than it needs to be. Also, adding traits that don't add much to the DL model will make it more time and space-consuming and change how accurate the results are. The suggested method includes preparation, choosing features, and teaching the model. The suggested method uses SG Denoising for preprocessing. CFS and MIFS are used in feature selection. The last step is to use Cap-A-BiLSTM to train the model. When compared to two other common methods, the proposed solution works very well. The results show that a Cap-A-BiLSTM can make more accurate predictions than the current classification technique.