Multilayer perceptron neural network-genetic algorithm for modeling Nicotiana tabacum leaf quality

Abstract

The global industry of tobacco (Nicotiana tabacum L.) is a profitable one comprising various products, including cigars, cigarettes, chewing tobacco, and smokeless tobacco. The internal quality of the cigarettes is highly related to the chemical components of tobacco leaves and shreds. Blue mold severity (BMS), chlorophyll (Chl), total nitrogen (N), sugar (S), nicotine (Nt), chloride (Cl), and potassium (K) contents of tobacco leaves are linked to the flavor and taste of cigarette products. A precise analysis of the effects of these factors would open the door for improving farmer income in low- and middle-income countries. In this study, BMS, Chl, N, S, Nt, Cl, K, green weight (GW), dry weight (DW), and leaf quality of four cultivars, including Bergerac, Bell, Burly, and Basma, were evaluated during two growing seasons. Bell displayed the highest leaf quality in two growing seasons. Multiple linear regression, stepwise regression, ordinary least squares regression, partial least squares regression, principal component regression, and multilayer perceptron neural network-genetic algorithm (MLPNN-GA) were used for the prediction of tobacco leaf quality responding to BMS, Chl, N, S, Nt, Cl, K. MLPNN-GA models displayed higher prediction accuracy compared with the best regression model according to R2 for MLPNN-GA vs. regression models were: Bergerac; 1.00 vs. 0.82, Bell = 1.00 vs. 0.41, Burly = 1.00 vs. 0.68, Basma = 0.94 vs. 0.68, and all cultivars = 0.94 vs. 0.66. The close match between the predicted and actual data validated the superior efficiency of the developed MLNNP-GA models for predicting tobacco leaf quality responding to BMS, Chl, N, S, Nt, Cl, K. Analysis of the developed MLPNN-GA models showed that Bergerac, Bell, Burly, and Basma leaf quality was most sensitive to BMS. MLPNN-GA was demonstrated to be a practical mathematical tool for predicting tobacco leaf quality in response to its chemical components and BMS.