Soil Fertility Prediction and Fertilizer Recommendation Using Deep Learning Approaches

Abstract

Agriculture has a significant role in defining a country's overall development, both economically and in terms of the quality of life for its citizens. We are now adopting smart farming approaches that use machine learning and data science to simplify the agricultural process and assist increase in both quantity and quality. The soil is an essential component of agricultural production because it contains the nutrients required to grow crops and adequate usage of appropriate fertilizer is a key for agricultural sustainability. So, we must first understand the properties and characteristics of various soil types to state the fertility status of the given soil samples and also type of fertilizer to apply. Therefore, this study is aimed to build predictive models for the prediction of soil fertility and suitable fertilizer by using a dataset collected from the Debre Markos soil laboratory and Adet agricultural research center which include 12,509 soil samples with 17 variables and 2 of them are dependent variables. Before feeding the data into the model it undergoes several preprocessing steps such as cleaning, handling missing and categorical values and normalization. Deep learning techniques such as Convolutional Neural Network (CNN), Artificial Neural Network (ANN), Deep Neural Network (DNN) and also Random Forest (RF) which is most previously used machine learning model is experimented in this study and results are compared to identify the most performing model. The performance of the models is evaluated under stratified 10-fold cross validation with performance metrics such as accuracy, precision, recall, and f1-score also confusion matrix is used to assess the classification report. And the performance evaluation results revealed that the CNN model best performs than other models for both soil fertility and fertilizer prediction since it scored an accuracy of 98.82%, precision of 98.80%, recall of 98.77%, and f1- score of 98.78% for soil fertility prediction and accuracy of 97.22%, precision of 98.56%, recall of 97.72%, and f1-score of 97.77% for fertilizer prediction. While this study uses deep learning techniques to predict soil fertility and fertilizer, it would be preferable to include soil nutrient level prediction by using a large dataset.