Leaf Classifications using Machine Learning

Abstract

In the field of plant ecological research, plant identification is a crucial step. An automatic leaf recognition approach for identifying plants is presented in this research. The suggested method is easy to use and effective in terms of calculation. This leaf dataset contains 340 rows, 16 attributes and 36 species. The dataset was downloaded from the UCI repository. We aim to classify the different species of leaf in the dataset using machine learning algorithms. The dataset used in this project consists of different species of the leaves that includes their specimen number, entropy, eccentricity, solidity, lobedness etc. Machine learning models like Decision Tree, Logistics Regressions, Support Vector Machine, XGBoost and AdaBoost were applied to this dataset. The accuracy, precision, F1 score and recall were calculated to assess the performance of the model. These results have implications for various learning and knowledge about different techniques. After assessing the performance of various algorithms on the raw data (without feature reduction), we observed that the performance of all algorithms were relatively low. Decision Tree and XGBoost performed comparatively well with an accuracy of 42% and 64% respectively. After applying the feature reduction using information gain and correlation matrix, we observed that the few attributes are highly correlated with each other. Hence, we removed three attributes(i.e ‘Third moment', 'Solidity', 'Average Intensity'). After performing feature reduction we see an incremental improvement in accuracy by 20–35% in Support Vector Machine, 40–70% in logistic regression, 70–90% in decision tree classifier, 75–99% in XGBoost and 25–50% in AdaBoost.