Areca Nut Disease Dataset Creation and Validation using Machine Learning Techniques based on Weather Parameters

Abstract

Areca nut crop yield is affected by many diseases caused due to heavy rainfall and high relative humidity. An early prediction of crop disease based on weather data can help farmers to take preventive measures. Many machine learning applications are deployed to detect the disease through image data. The proposed study is the first approach for creating a novel dataset and developing the weather-based areca nut disease prediction model. Historical weather data i.e. temperature, rainfall, relative humidity, sunshine, wind direction, and wind speed are collected from the Udupi weather station. Fruit rot disease data are collected through farmer surveys, disease management recommendations, and research literature. These data are integrated and correlated to create the final dataset which is validated and compared using a statistical method, decision tree regression (DTR), multilayer perceptron regression (MLPR), random forest regression (RFR), and support vector regression (SVR) models. Principle component analysis, branch and bound, and wrapper feature selection techniques are used to select the weather parameters contributing to more accurate prediction. The observation shows that RFR gives 0.9 mean absolute error (MAE) as the lowest value among many models and SVR gives 1.7 MAE as the highest error after feature selection.