Influence of fuzzified dataset on classification and prediction of plant types - A case study

Abstract

This research explores the use of fuzzification to improve the classification and prediction of plant types based on environmental and soil parameters. Fuzzification, a process that transforms numerical features into fuzzy sets, is used to handle the inherent uncertainty discovered in parameters such as soil pH, moisture, nutrients and temperature. The dataset obtained from Kaggle consists of 9 features and 10 plant types. Several Machine Learning models such as Naïve Bayes, Support Vector Machine, Random Forest, K-Nearest Neighbour, Decision tree, XGBoost and LightGBM are employed to classify and predict plants based on their environmental and soil features. These models are applied to fuzzified and non-fuzzified datasets for comparative performance analysis. The hyperparameters of each model is fine-tuned using the Bayesian optimization. SVM and KNN significantly benefit from the fuzzified dataset demonstrating the effect of fuzzification. XGBoost achieves an accuracy of 91.37% and AUC of 99.41% on the fuzzified dataset, while with the non-fuzzified dataset, accuracy and AUC of 91.34% and 99.42% respectively is found to be achieved. LightGBM shows an accuracy of 91.35% and AUC of 99.41% on the fuzzified dataset and 91.27% accuracy and AUC of 99.40% on the non-fuzzified dataset. From this research work, fuzzification is observed to improve the ability of certain models to manage complex data, leading to more accurate classification. These findings aid in the enhancement of more reliable and robust machine learning models for agricultural applications, particularly in prediction and management based on uncertain environmental and soil parameters.