Adaptive Apriori Algorithm 🚀

Overview

This project introduces an optimized version of the Apriori Algorithm for Association Rule Mining. Instead of relying on arbitrary, manually-set parameters, this implementation uses Dynamic Thresholding to automatically calculate the optimal Minimum Support ($MinSup$) and Minimum Confidence ($MinConf$) based on the statistical distribution of the dataset.

Key Features

• Dynamic Parameter Calculation: Uses the median and standard deviation ($\sigma$) of item frequencies to set thresholds tailored to the dataset's density.
• Adaptive Level-Wise Approach: Recalculates thresholds at each level ($k=1, 2, 3...$) to prioritize high-quality rules and aggressively prune the search space.
• High Performance: Significant reduction in computational overhead, making it ideal for large or dense datasets.
• Medical Insights: Tested on heart disease data to identify significant correlations between symptoms and diagnoses.

📊 Performance Benchmarks

Tested on the heart_nominal.csv dataset (988 transactions):

Metric	Classic Apriori	Dynamic Hybrid	Adaptive Version
Patterns Found	165	49	22
Rules Generated	107	15	8
Avg. Support	0.249	0.371	0.402
Execution Time	4.878s	0.492s	0.124s

Result: The Adaptive version achieved a 97.4% speed increase compared to the classic approach.

Technologies Used

• Python 3.12
• Pandas & NumPy (Data processing)
• Matplotlib (Data visualization)
• Itertools (Optimized combination generation)

How It Works

1. Initial Scan: Computes the support of all individual items.
2. Statistical Analysis: Calculates the median and standard deviation of those supports.
3. Adaptive Pruning: Sets a high $MinSup$ for the initial levels to filter out "noise" and lowers it as itemsets become more complex, ensuring only the most relevant patterns are explored.

Author

**Ferhat Abderrahmane ** Student at USTHB (University of Science and Technology Houari Boumediene).