Predictive Soil Analytics Using Data Mining Techniques

Abstract

Agriculture and allied industries play an important role in the development of our nation. In India more than 55% of people make a living from farming. Crop yields are an essential aspect of every farmer’s day. It depends on many factors like soil quality, seeds, planting practices, humidity, fertilizers and pesticides. Besides all factors, diagnosing soil quality is a fundamental and essential task in farming, as it provides background knowledge of the soil and its physical, chemical and biological prominence.  Hence, soil analytics is inevitable that gives information about the present nutrient availability or the need of the nutrients for effective cultivation. It helps to interpret the physico-chemical properties of soil nutrients and to classify the nutrient content as very low, low, medium, high, or very high based on pH values. Thus, predictive analytics based on the soil parameters offer precise and sensible solutions for soil fertility problems and enable suitable decision on crop cultivation. This study attempts to exploit the benchmark classification algorithms from data mining to classify soil samples of Tiruppur district using pH levels. The prediction of pH levels is important to know the nutrients availability in the soil. Classification algorithms like Logistic Regression (LR), Bernoulli Naïve Bayes (BNB), Decision Tree (DT), Extra Tree (ET), Random Forest (RF) and K-Nearest Neighbor (KNN) are used to evaluate and predict the pH values. After the comprehensive evaluation, this study determined that the performance of the DT and RF model for pH prediction is high compared to the other algorithms in terms of accuracy. Further, the classifiers performance has improved by possessing feature scaling techniques like normalization and standardization. Results showed that the prediction accuracy of KNN and BNB with feature scaling outperforms the other algorithms.