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

Area of Science:

• Agricultural Science
• Plant Science
• Data Science

Background:

• Tobacco (Nicotiana tabacum L.) quality directly impacts cigarette flavor and industry profitability.
• Key internal quality factors include blue mold severity (BMS), chlorophyll (Chl), nitrogen (N), sugar (S), nicotine (Nt), chloride (Cl), and potassium (K).
• Accurate prediction of these factors can enhance farmer income, particularly in low- and middle-income countries.

Purpose of the Study:

• To evaluate the leaf quality of four tobacco cultivars (Bergerac, Bell, Burly, Basma) under varying conditions.
• To develop and compare predictive models for tobacco leaf quality based on chemical composition and BMS.
• To identify the most influential factors affecting tobacco leaf quality.

Main Methods:

• Field evaluation of BMS, Chl, N, S, Nt, Cl, K, green weight (GW), dry weight (DW), and leaf quality over two growing seasons.
• Application of multiple linear regression, stepwise regression, ordinary least squares, partial least squares, principal component regression, and MLPNN-GA for quality prediction.
• Comparative analysis of model accuracy using R-squared values.

Main Results:

• Bell cultivar exhibited the highest leaf quality across both growing seasons.
• MLPNN-GA models demonstrated superior prediction accuracy (R2 up to 1.00) compared to the best regression models (R2 up to 0.82).
• Analysis revealed BMS as the most significant factor influencing the leaf quality of all tested cultivars.

Conclusions:

• MLPNN-GA is a highly effective tool for predicting tobacco leaf quality based on chemical components and BMS.
• The developed models offer a practical approach to optimizing tobacco quality and potentially increasing farmer profitability.
• Understanding the sensitivity of leaf quality to specific factors like BMS is key for targeted agricultural management.