Fragile Watermarking of Decision System Using Rough Set Theory

The wheels of modern society are driven by multitudes of databases that serve as repositories of valuable information. Several applications can fruitfully learn from special repositories called decision systems. DS are databases that contain records of objects described by condition attributes and labeled by decision attributes that categorize them into distinct classes. Rough set theory can be applied to derive high quality classification rules from them to predict accurate decisions on freshly gathered data. For security, it is necessary to protect this core information from vulnerabilities in the internet. No prior work is done on watermarked protection of decision systems. To fulfill this gap, we propose a new fragile and blind watermarking scheme for tamper detection in decision systems which detects even the slightest integrity losses that may damage the classificatory information encapsulated within a decision system. The technique characterizes the type of attack and then localizes the perturbation up to an attribute's value level. In case of alteration in reducts, proposed technique can recover the original value. The watermarking technique first prepares secure signature by encoding the information on reducts, rules and their support values. It securely embeds this into dataset. We present theoretical analysis to illustrate the high degree of fragility of our proposed scheme to different kinds of attacks. Experimental results demonstrate that with the modification of up to 50% addition, deletion or alteration of tuples, the watermark recreated from the compromised database reflects the changes close to 50% on an average, thus facilitating immediate detection.