Volume 18, No. 6, 2021

Enhanced Feature Engineering By Discretized Naïve Bayes To Predict Soil Fertility For The Betterment Of Sugarcane Yield

Raynukaazhakarsamy , Dr. J.G.R. Sathiaseelan


The challenging task of feature engineering in data mining is the dimensionality reduction to extract relevant attributes. Hence, the inherent analysis of data distribution based on its class label is essential for predicting the results. This research work contributes two machine learning models namely Wrapper based Discretized Naïve Bayes (WDN Bayes) and Filter based Discretized Naïve Bayes (FDN Bayes) using the supervised machine learning algorithms such as NB, KNN and DN Bayes that aims for identifying an optimal subclass of features from the collection of primary soil dataset based on chemical nutrients around Theni region to predict soil fertility by improving the classification accuracy. Extensive experiments on four different real-time soil datasets were carried over to demonstrate the effectiveness of KNN embedded wrapper method and CFS+GA combined filter method using Discretized Naïve Bayes (DN Bayes). The model’s effectiveness is estimated with all the features and with the significant features obtained by the proposed feature extraction techniques. The experimental results of feature extraction approach profoundly satisfying in terms of error metrics and evaluation metrics in comparison with NB, KNN, SVM, DN Bayes, WDN Bayes and FDN Bayes by producing 91% and 92% of classification accuracy for WDN Bayes and FDN Bayes respectively.

Pages: 2876-2896

Keywords: Soil fertility, Crop yield Prediction, Soil Classification accuracy, Correlation filter, Wrapper approach and Feature extraction.

Full Text