Volume 18, No. 6, 2021

Outlier Detection And Denoising By The Measures Of Dispersion With Naïve Bayes To Predict Soil Fertility

Raynukaazhakarsamy , Dr. J. G. R. Sathiaseelan


Farming is one of the essential segments of Indian Economy and it funds about 17% of Gross Domestic Product (GDB). Growing crops more than millennia without thinking often about renewing has prompted consumption and depletion of soil supplements bringing about their low usefulness. To work on production, compound composts are added. Adequate measure of manures should be added for the improvement of good yield and simultaneously the normal nature of soil stays perfect. The soil should be identified as fertile or not fertile to carry out the process of manuring. Methods/Statistical analysis: It is essential to remove redundant data, replace missing values and outliers from the soil dataset. Removing duplicates and replacing missing values are performed using Discretized Naïve Bayes (DNBayes) and Outliers are detected and replaced using Discretized Naïve Bayes with Quartiles (DNBQ). Findings: The performance of the proposed system is analyzed in terms of different types of errors like KS, MAE, RMSE, RAE and RRSE along with TPR, FPR, Precision, Recall, F1- Score, ROC and classification accuracy. Prediction of soil fertility based on DNBayes and DNBQ achieved 85% and 87% of classification accuracy. Novelty/Applications: The proposed approaches will benefit soil scientists in decision making to help farmers in predicting fertile soil for sugarcane cultivation.

Pages: 2140-2156

Keywords: Discretized Naïve Bayes; Denoising; Outlier Detection; Inter Quartile Range; Soil Fertility.

Full Text