Effect of outliers in statistical modelling for predicting the outbreak of anthracnose in grapes (Vitis vinifera)

Abstract

Role of outliers while developing statistical models for disease prediction is studied. Specifically, statistical models were developed to optimize the role of weather factors and simultaneously to predict the incidence of anthracnose in grapes (Vitis vinifera) by eliminating the effect of aberrant /outlier data. It was observed that about 93.3% of the disease incidence was collectively explained by weather factors with a time lag of one week prior to incidence as expressed by the equation (Y = –86.87 +8.62 max.temp –0.98 min.temp –1.4 RH1 +0.64 RH2+6.9 ws–8.6 Evap+0.07 Rf–0.08 no.rd–0.6 Br.SS–9.2 VPD ) of the variability in PDI. Out of this, two factors namely, wind speed and vapor pressure deficit could explain about 80.1 % of the anthracnose incidence, as explained by the equation Y=5.2+5.95ws–8.1VPD. It was noted that effect of 13 outliers/ aberrant observations when removed increased the prediction power of the model by 14 %. Moreover, as a measure of goodness-of-fit, the coefficient of determination (R2) was used to evaluate the empirical models developed. Before taking final conclusion about the model, the model-generated residuals were tested for their robustness using statistical techniques.