The rapidly evolving field of smart agriculture has made farming practices much faster, simpler and easier, and helped perform complex tasks like yield prediction, crop disease detection, crop data analysis, etc. with ease. However, factors such as the cost that comes with the implementation and maintenance of smart agriculture systems has been discouraging farmers from adopting it entirely. This study aims to reduce the cost of implementation, focusing on one area, which is the reduction of the number of sensors to be employed for data collection. The best substitutions using only 2–3 sensors have been found to replace common sensors. The sensors for each replacement have been determined by finding out the relevance of various sensors in prediction using Feature Importance, Permutation Importance and employing XAI. Multiple Machine Learning and Deep Learning algorithms like Support Vector Machine (SVM), Random Forest, XGBoost, Artificial Neural Network (ANN) have been trained to predict the outputs. The substitutions have been compared on the basis of the Mean Squared Error given by the particular predictive model. A web application has also been developed using Streamlit, which allows users to obtain the predicted values of sensors based on the data from the sensors available to them. Overall, this paper demonstrates the effectiveness of Machine Learning in reducing costs in the field of smart agriculture.