Article Preview
Top1. Introduction
Machine learning in agriculture is an attractive novel research area. Agriculture data are extremely diversified in terms of environment, climate, weather condition, interdependency, and use of sources for farming. The main problem of using Machine Learning in agriculture is to solve issues based on the available data and its meaningful outcomes (Patel & Kathiriya, 2017). Agriculture plays a vital role in the Indian economy and the production of crops. Crops may be either commercial crops or food crops. Food crops include rice paddy, maize, wheat, grams, millets, etc., while commercial crops are cotton, sugarcane, groundnut, cashew, etc. The productivity of the crops is drastically influenced by weather conditions (Palanivel & Surianarayanan, 2019). It is necessary to cultivate the soil properly for maintaining fertility, achieve better yield and protect the environment. Present challenges of the agriculture domain are uncertain climatic changes, poor irrigation facilities, weather uncertainty, water shortages, reduction in soil fertility, and uncontrolled cost due to demand-supply impose farmers to be equipped with smart farming. A soil test is the learning of a soil sample to ascertain an additional matter, its composition, and various attributes. Generally, soil tests are accomplished to determine the wealth (Bhuyar, 2014). Ensemble methods are meta-algorithms that combine several machine learning techniques into predictive models like bagging to decrease variance, boosting to reduce bias, and stacking for improving predictions. This article presents the evaluation of three ensemble classifiers on the soil datasets including bagging, boosting, and stacking. These classifiers are tested to achieve the best accuracy of the dataset. Bagging is an ensemble classifier that creates separate samples of the training datasets and creates a classifier for each sample. The results of these multiple classifiers are then combined such as averaged or majority voting. Boosting is an ensemble method that starts with a base classifier that is arranged on the training data. The second classifier is created behind the first classifier to focus on the instances in the training data. The process continues to append classifiers in anticipation of a limit is reached in the number of models or accuracy. The multiple different algorithms are prepared on the training data and a meta-classifier is prepared, which learns how to take the predictions of each classifier and make accurate predictions on unknown data as stacking. This research work utilizes the sample data that have been collected from the Vellore soil testing laboratory. In the beginning, the data has been preprocessed and then appropriate features are extracted using the feature extraction method then passed onto the various training ensemble classifiers of machine learning to acquire a better result.
1.1 Objectives
- 1.
Investigating the important factors and selecting the high dominant features that affect the prediction of soil fertility.
- 2.
Identifying the key factors that associate with Soil fertility.
- 3.
Developing models to classify the fertility as Ideal or Not Ideal.
1.2 Outline of the Paper
This proposed work has been organized as follows: Section 2 describes the characteristics of soil and soil fertility. Related works on Machine learning and ensemble classifiers are explained in Section 3. Section 4 gives the Materials and Methods used. The Proposed methodology with the research framework is outlined in section 5. Section 6 describes Experimental results and discussion, and Section 7 concludes research work with future directions.