The Implementation of Machine Learning and Deep Learning Algorithms for Crop Yield Prediction in Agriculture

Keywords: machine learning, deep learning, MR, MLR, DNN,GBRT, gradient descent. 
Introduction. Artificial intelligence (AI) has 
become a crucial technology in the Fourth 
Industrial 
Revolution, 
gaining 
substantial 
recognition in various domains such as finance, 
healthcare, and manufacturing. The subfields of AI, 
specifically machine learning and deep learning, 
have become widely prevalent in diverse areas, 
including speech recognition, computer vision, 
language models, and industrial fault diagnosis [1]. 
Consequently, AI has attracted significant attention 
as a revolutionary force capable of driving these 
fields forward and enhancing human abilities, 
presenting enormous potential for industry 
transformation. 
Given the critical role of agriculture in the 
global economy, understanding global crop yield 
patterns is essential for addressing food security 
challenges and mitigating the impact of climate 
change amid a growing human population. 
Accurately predicting crop yields is a significant 
agricultural challenge that depends on multiple 
factors such as weather conditions (e.g., rainfall, 
temperature) and pesticide application. Therefore, 
having precise knowledge of crop yield history is 
crucial when making decisions related to 
agricultural risk management and yield forecasting 
[1][2]. Crop yield prediction poses a challenge for 
decision-makers at various levels, from global to 
local scales. Farmers, for instance, can leverage 
reliable crop yield prediction models to determine 
optimal planting schedules and crop selection. 
There are various approaches to forecasting crop 

---

Bulletin of TUIT: Management and Communication Technologies
Nodir Rahimov, Dilmurod Khasanov 
2023.Vol-2(4) 
yields [2]. 
Machine learning represents a practical 
approach that can facilitate improved crop yield 
prediction by leveraging multiple attributes. As a 
subdivision of Artificial Intelligence (AI) that 
emphasizes learning, machine learning (ML) is 
capable of extracting insights from datasets by 
identifying correlations and patterns. During the 
training phase, ML models are trained using 
datasets that capture prior experiential outcomes, 
and the resulting predictive models incorporate a 
range of features and parameters calculated from 
previous data. During the testing phase, unused 
historical data is employed to evaluate model 
performance. Depending on the research question 
and topic, ML models can be descriptive or 
predictive. Predictive models leverage past data to 
forecast future events, while descriptive models 
help to characterize current conditions or historical 
trends. Machine learning techniques have been 
instrumental in improving crop yield prediction and 
crop management decision-making. In recent years, 
a range of machine learning algorithms such as 
multivariate regression, decision trees, association 
rule mining, and artificial neural networks have 
been deployed to enhance crop yield forecasting in 
agriculture [5] [6]. 
This paper's primary contributions are as follows: 
1. The primary objective of this study is to 
compare the performance of various 
machine learning and deep learning models, 
including the multivariate regression (MR), 
deep neural network (DNN), and multiple 
linear regression (MLR), in predicting the 
wheat yield for the next season using 
previously collected data. 
2. We provide a detailed description of the 
data preprocessing procedure used in our 
study. This process includes importing raw 
data and constructing a dataset that includes 
soil moisture, rainfall, temperature, and 
volume of minerals (Nitrogen, Phosphorus, 
natural minerals). Additionally, we remove 
unnecessary data and classify the 
specifications for model training and 
validation. 
3. To validate and evaluate the effectiveness 
of our study, we conduct a comprehensive 
comparative analysis of various models 
used to predict crop yield for the next 
season. This analysis includes an 
assessment of the accuracy of these models, 
allowing us to determine which approach is 
most effective in accurately predicting crop 
yield.