A Suitable Crop Recommendation using Multiple Kernel Feature Selection Based on Sigmoid Activation in Recurrent Neural Network

Abstract

Agriculture plays a major role in economic development. The main difficulties present among the farmers are they are not selecting the right crop based on the present weather and soil conditions. They are simply following the ancestral farming patterns, which may sometimes lead to the wrong selection of crops that affects productivity. This problem of the farmers has been addressed by predicting the most suitable crop based on the soil condition and forecasted weather in a particular region. The proposed system aggregates soil, crop, and weather data. By applying Multiple Kernel Feature Selection based on sigmoid activation in Recurrent Neural Network (MKFS-SARNN) the most suitable crop is predicted based on the current environmental condition. Initially, the proposed system collects the data and performs feature selection using a Multiple Kernel Fuzzification model. The features get seeded into fuzzy rules make mean weightage along with marginal dataset from rainfall, temperature, humidity, etc. the Fuzzified weight is fed into Recursive Feature wrapping model (RFWM) to make spatial crop information data rate (SCIDR). Further, the selected features are fed into the sigmoid activation in Recurrent Neural Network. The classifier is trained and tested with the selected features which categorize the data into recommending and non-recommend fields. The proposed system produces more accurate results in recommending crops that increase productivity which reduces the wrong choice on a crop selection and increase in productivity.

 Keywords:  Crop recommendation, weather forecasting, Kernel function, Fuzzy logic, SIDR rate, Recurrent Neural Network, feature selection.