Record Details

SOME INVESTIGATIONS ON DIFFERENT CLASSIFICATION TECHNIQUES IN AGRICULTURE

KrishiKosh

View Archive Info
 
 
Field Value
 
Title SOME INVESTIGATIONS ON DIFFERENT CLASSIFICATION TECHNIQUES IN AGRICULTURE
 
Creator SAMARENDRA DAS
 
Contributor Amrit Kumar Paul
 
Subject Service failed, response=Internal Server Error
 
Description T-8496
Classification is of broad interest in science because it leads to many scientific studies and also arises in the contexts of many applications. For example in agriculture, crop varieties are classified in to different groups which are suitable for different agroclimatic zones of a region. Classification techniques are the multivariate techniques which are based on some assumptions. But in crop data situation, there may be violation of assumptions of the classification techniques. Hence, the purpose of the study to investigate how the classification techniques perform when certain assumptions about the data characteristics are violated. Further, occurrence of missing observations in multivariate crop data is a common feature in agricultural experiments due to the non-response of genotypes, disease and pest attack. The Classification of genotypes in presence of missing values is a challenging task for breeders. So, the present investigation has been taken up to achieve the afore mentioned goal with the following objectives: (i) To compare empirically the performance of Oblique Axes, k-th nearest neighbor, Linear and Quadratic Discriminant procedures under multivariate skew-normal situations. (ii) To assess the performance of different imputation techniques against missing observations for different classification procedures. The performance of different classification techniques viz. Oblique Axes Method (OAM), k-th nearest neighbor (KNN), Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA) are compared based on Apparent classification Error Rate (APER). Further the result shows that, KNN followed by OAM and LDA perform better in skew-normal situations than normal condition and QDA performs better in normal condition. For maximum consistency and accuracy of classification of skew-normal data, KNN is best among the above four classification techniques. The performance of the above four classification techniques are also studied under 1%, 5%, 10% and 20% missing observations created randomly in the original data, which are imputed by different methods like zero, mean, regression and multiple imputation methods based on the weighted average Hit Ratios. The present results reveal that all the imputation methods are robust against 1% and 5% missing observations. Further, it is found that mean, regression and multiple imputation techniques performs well in case of 10% missing observations. In case of 20% or more missing observations the regression and multiple imputation provide better results. Among four the classification techniques, KNN technique performs best irrespective to the different levels of missing observations.
 
Date 2016-11-28T14:25:54Z
2016-11-28T14:25:54Z
2011
 
Type Thesis
 
Identifier http://krishikosh.egranth.ac.in/handle/1/87869
 
Format application/pdf
 
Publisher IARI, Indian Agricultural Statistics Research Institute