Code-Readability-Analysis

This project implements a readability analysis framework for Java code snippets, combining software metrics with machine learning classification.

---

📌 Overview

The tool analyzes Java code snippets (.jsnp files) and extracts several well-known software metrics that correlate with readability.
It then builds a labeled dataset using human readability scores and applies a machine learning classifier to predict readability.

---

✨ Features

The following readability metrics are implemented:

1. Number of Lines (LOC)
   Counts the total number of lines in a snippet (including comments and blanks).

2. Halstead Volume

   • Derived from the number of distinct and total operators (n1, N1) and operands (n2, N2).
   • Formula:
     [ V = (N1 + N2) \times \log_2(n1 + n2) ]

3. Cyclomatic Complexity

   • Starts at 1, and increments for each control flow element:
     if, for, while, do-while, foreach, switch cases, catch blocks, ternary ?:, and logical connectors (&&, ||).

4. Token Entropy

   • Uses Shannon entropy over token distributions (including whitespace and special symbols).
   • Captures predictability vs. diversity of tokens.

---

⚙️ How It Works

1. Preprocessing (Dataset Creation)

   • Extracts the selected feature metrics from .jsnp files.
   • Reads a ground truth CSV containing human readability ratings.
   • Converts ratings into binary labels:
     • Average score ≥ 3.6 → Y (Readable)
     • Average score < 3.6 → N (Not Readable)
   • Saves the output as a CSV dataset.

2. Classification

   • Loads the dataset with Weka.
   • Trains a Logistic Regression classifier.
   • Evaluates via 10-fold cross-validation.
   • Reports Accuracy, Area Under ROC, and F-Score.

---

🚀 Usage

The project is packaged as a runnable JAR with subcommands (using Picocli).

Preprocess – build dataset

java -jar Readability-Analysis-1.0.jar preprocess \
  --source ./snippets \
  --ground-truth ./truth.csv \
  --target ./dataset.csv \
  lines h_volume token_entropy cyclomatic_complexity