ML-BASED BLOCK PARTITIONING FOR ADAPTIVE HUFFMAN CODING

This project introduces an adaptive Huffman coding method that uses machine learning to guide block partitioning during lossless compression. It extracts statistical features like byte frequency distribution, Shannon entropy, run-length statistics, and local symbol variation over sliding windows of the input stream to detect changes in data characteristics. A simple supervised model, trained with entropy-based heuristic labels, scores potential positions as block boundaries. After identifying these boundaries, adaptive Huffman coding is applied independently within each selected block. Limits on minimum and maximum block size, along with a fallback heuristic segmentation, ensure reliable compression, even when the model lacks confidence. Experimental tests across diverse datasets, including log files, source code, sensor readings, and mixed text, show that this machine learning-guided block partitioning improves the compression ratio and space savings compared to fixed-size and purely heuristic block schemes. It maintains correct decompression and keeps runtime overhead within acceptable limits.