Applying Some Statistical Methods in WEKA

Statistical analysis and evaluation of machine learning results

Dr. Jamolbek Mattiev

Role of Statistics in WEKA

• Statistics help evaluate, compare, and validate models
• Used to measure performance, reliability, and significance
• WEKA integrates statistical methods with machine learning

Types of Statistical Methods in WEKA

WEKA supports statistics for:

• Descriptive analysis
• Performance evaluation
• Algorithm comparison
• Feature selection

Descriptive Statistics (Preprocess Tab)

Available in Explorer → Preprocess:

• Mean
• Standard deviation
• Minimum and maximum
• Distribution of attributes

​

Purpose: Understand data before modeling

Data Visualization for Statistical Insight

WEKA provides:

• Histograms
• Scatter plots
• Class-wise data visualization

Used to:

• Detect outliers
• Understand class separation
• Identify correlations

Statistical Evaluation Metrics

After classification, WEKA reports:

• Accuracy
• Precision
• Recall (Sensitivity)
• F-measure
• ROC Area (AUC)

These are statistical measures of performance.

Confusion Matrix Analysis

WEKA generates a confusion matrix:

• True Positives (TP)
• False Positives (FP)
• True Negatives (TN)
• False Negatives (FN)

Used to compute many statistical metrics.

Cross-Validation as a Statistical Method

• WEKA commonly uses k-fold cross-validation
• Data is split into k subsets
• Results are averaged statistically

​

Purpose: Reduce bias and variance

Experimenter: Statistical Comparison of Algorithms

• WEKA Experimenter allows:
• Running multiple algorithms
• Comparing results statistically
• Reproducible experiments

WEKA supports:

• Paired t-test
• Corrected resampled t-test

Used to determine whether performance differences are statistically significant.

Feature Selection and Statistics

In Select Attributes tab:

• Information Gain
• Chi-Square test
• Correlation-based methods

These are statistical feature evaluation techniques.

Example: Applying Statistics in WEKA

Typical workflow:

• Load dataset
• Explore descriptive statistics
• Apply classifier
• Analyze statistical metrics
• Compare algorithms using Experimenter

Benefits of Statistical Analysis in WEKA

• Objective model evaluation
• Reliable algorithm comparison
• Better understanding of data behavior
• Scientifically sound conclusions

Limitations

• Limited customization of tests
• Assumptions (normality, independence)
• Not a full statistical software

Summary

• Statistical methods are essential in ML analysis
• WEKA integrates statistics with modeling
• Proper statistical evaluation improves research quality