ClassifiCation analysis using
orange Data Mining tool
By:
MohaMeD ashik Basha M
urk22eC6007
Contents
1 Introduction 2
1.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Dataset Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Methodology 4
2.1 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Classification Models Used . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Results and Analysis 6
3.1 Visualization of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Performance Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4 Conclusion 8
1
1
Chapter 1
Introduction
1.1 Problem Statement
The primary objective of this report is to perform a classification analysis on a selected
dataset using the Orange Data Mining Tool. Classification is a supervised machine learn- ing
technique used to predict the category or class of new observations based on training data. In
this analysis, we aim to load a dataset, preprocess it, build multiple classification models,
evaluate their performance using visualization tools like confusion matrices and ROC curves,
and compare them based on key metrics such as accuracy, precision, recall, and F1-score.
This exercise demonstrates the ease and effectiveness of visual program- ming in Orange for
machine learning tasks, particularly for educational and exploratory purposes.
The problem can be framed as: Given a dataset with features and target class labels,
develop and compare classifiers to achieve the highest predictive performance. We select the
Iris dataset for its simplicity and classic nature in classification problems.
1.2 Dataset Description
The Iris dataset, originally introduced by Ronald Fisher in 1936, is a multivariate dataset
used for classification tasks. It consists of 150 samples from three species of Iris flowers: Iris
setosa, Iris versicolor, and Iris virginica. Each sample is described by four features:
• Sepal length (in cm)
• Sepal width (in cm)
• Petal length (in cm)
• Petal width (in cm)
The dataset is balanced with 50 samples per class and has no missing values, making it
ideal for initial preprocessing and modeling. In Orange, the dataset is loaded via the File
widget, which reads the CSV format of the Iris data available in UCI Machine Learning
Repository.
2
Figure 1.1: Sample of the Iris Dataset Loaded in Orange
9
Chapter 2
Methodology
2.1 Preprocessing
Preprocessing in Orange is handled visually using widgets like Continuity for data type
handling and Preprocess for transformations. For the Iris dataset:
• Missing Values: No missing values were present, so no imputation was needed.
• Normalization/Standardization: Features were standardized using the Normalize
widget with z-score normalization to ensure all features contribute equally to the
models, especially for distance-based algorithms like k-NN.
The preprocessed data is then connected to the Test and Score widget for model
validation using stratified cross-validation (10-fold).
2.2 Classification Models Used
We implemented three classification algorithms in Orange to provide a diverse compari- son:
Decision Tree, Random Forest, and k-Nearest Neighbors (k-NN). These were selected for their
representation of different paradigms—tree-based, ensemble, and instance-based learning.
• Decision Tree: Uses the Tree widget with entropy as the splitting criterion. It
builds a hierarchical structure of decisions based on feature thresholds.
• Random Forest: An ensemble of 10 decision trees via the Random Forest widget,
reducing overfitting through bagging and feature randomness.
• k-Nearest Neighbors (k-NN): Implemented with the kNN widget, using k=5
neighbors and Euclidean distance metric on standardized features.
Models were trained and evaluated using the same cross-validation setup in the Test and
Score widget.
10
Figure 2.1: Orange Workflow for Classification Pipeline
This illustration depicts the visual workflow in Orange, connecting File → Preprocess
→ Models → Test and Score → Visualizations.
9
Chapter 3
Results and Analysis
3.1 Visualization of Results
Results were visualized using the Confusion Matrix and ROC Analysis widgets in Or- ange.
Figure 3.1: Confusion Matrix for Random Forest Model
The confusion matrix illustrates prediction accuracies across classes, showing minimal
misclassifications for setosa and higher errors between versicolor and virginica due to feature
overlap.
10
Figure 3.2: ROC Curves for All Models
ROC curves plot true positive rate vs. false positive rate, with AUC values indicating
overall performance (closer to 1 is better).
3.2 Performance Comparison
Model performances were compared using the following metrics from the Test and Score
widget:
Model Accuracy Precision Recall F1-Score
Decision Tree 0.95 0.95 0.95 0.95
Random Forest 0.97 0.97 0.97 0.97
k-NN 0.96 0.96 0.96 0.96
Table 3.1: Performance Metrics Comparison
Random Forest achieved the highest scores across all metrics, demonstrating robust- ness.
Decision Tree was simplest but prone to slight overfitting, while k-NN benefited from
standardization.
9
Chapter 4
Conclusion
In this classification analysis using Orange, the Random Forest model outperformed De- cision
Tree and k-NN with an accuracy of 0.97, attributed to its ensemble nature that averages multiple
trees to reduce variance. Orange’s drag-and-drop interface facilitated rapid experimentation and
visualization, making it suitable for beginners. Future work could explore hyperparameter
tuning or additional datasets like Titanic for imbalanced classes.
10