Identification of causal relationships among clinical variables for cancer diagnosis using multi-tenancy

Cancer causing more deaths than AIDS, tuberculosis and malaria combined. Especially breast cancer killing more than 40,000 women and 440 men every year in U.S.A. Over many years various data mining studies have tried to predict the cancer. There are only few studies on finding causal relationship among clinical variables causing cancer. They also provide theoretical guidance for cancer diagnosis and treatment. As there are many classifiers, learners and techniques to find causal relationships, it is very difficult to find attributes with very strong positive relation that are causing cancer. In this paper, we have applied Multi-Tenancy strategy based on logical databases, where whole database is divided into four tenants and proposed a graphical structure of key-dependency attributes which are causing cancer. We have used Pearson Product Moment Correlation Coefficient (PPMCC) to measure the strength of linear relationship between attributes and kappa analysis for finding the efficiency of each tenant. The tenant with highest kappa measure is treated as more efficient tenant. The proposed algorithm applies searching algorithm on conditional mutual information matrix to identify attributes which are dependent. This method represents relationships between attributes by using directed acyclic graph. Thus instead of finding general relationships, it is very useful to find very strong positive relationships which improves the accuracy in diagnosing cancer causing attributes.