Development of Ensembled Probabilistic Machine Learning models on Agri-Climatic data
Loading...
![Thumbnail Image](assets/images/Item.jpg)
Date
2023-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
G. B. Pant University of Agriculture & Technology, Pantnagar-263145
Abstract
Mathematical models serve as formal representations of real-world phenomena,
employing mathematical equations, symbols, and logical relationships. They are
indispensable tools for describing, analyzing, and predicting the behaviour of complex
systems using mathematical principles and techniques. This study investigates
deterministic and probabilistic mathematical models, including Logistic Regression,
Naïve Bayes Classifier Models, Hidden Markov Models, Artificial Neural Network
Models with a SoftMax output layer, Decision Tree Models, and K Neighbors Models,
for rainfall prediction based on agri-climatic data. The models are trained using
secondary data obtained from the NASA POWER database, specifically the Agriclimatology
database related to Pantnagar, Uttarakhand. The dataset consists of 4,383
rows and 14 columns, representing daily data from 2010 to 2021. Model validation is
conducted using data from the year 2022, and performance evaluation utilizes standard
measures commonly employed in machine learning studies. The results indicate that the
probabilistic models, particularly Logistic Regression, Hidden Markov Models, Naïve
Bayes Classifier Models, and Artificial Neural Network Models with a SoftMax output
layer, exhibit promising performance for rainfall prediction. Ensemble models are
developed using averaging, max voting, and stacking techniques, effectively combining
multiple models to enhance prediction accuracy. Among the ensemble models, the
stacking model demonstrates the highest accuracy, followed by the Max Voting model
and the averaging model.
Description
Keywords
Citation
Theses of M.Sc.