Abstract:
Accurate forecasting of guava harvest is essential for efficient resource allocation, market
planning, and mitigating post-harvest losses. In Bangladesh, the guava industry faces
challenges in predicting harvest yields due to the complex interaction of various
environmental factors. This study proposes a novel approach to enhance guava harvest
forecasting in Bangladesh through the application of supervised ML models. The research
leverages historical guava production data and corresponding meteorological variables,
including temperature, humidity, precipitation, and solar radiation. These variables are
used as input features for training and testing several supervised ML models, such as linear
regression, decision trees, random forests, support vector machines, and artificial neural
networks. A comprehensive dataset comprising guava production records and
meteorological data from multiple regions in Bangladesh is collected and preprocessed.
Feature engineering techniques are employed to extract relevant information from the data
and optimize model performance. The dataset is then divided into training and testing sets
for model development and evaluation. Performance metrics such as MAE, RMSE, MSE
are used to assess the accuracy and reliability of the machine learning models. Where the
highest accuracy 84.72% is achieved by DTR. And the lowest accuracy is achieved by
LinR accuracy of 43.07%. The models' forecasting capabilities are compared, and the most
effective model is identified. The results demonstrate that the supervised machine learning
models exhibit promising performance in guava harvest forecasting, outperforming
traditional statistical methods. The selected model achieves high accuracy and provides
valuable insights into the influence of meteorological variables on guava production. The
findings of this study have significant implications for the guava industry in Bangladesh,
helping to enhance productivity, reduce wastage, and promote sustainable agricultural
practices. Moreover, the methodology presented are extended to other regions and crops,
facilitating improved harvest forecasting in diverse agricultural contexts.
