Intraoperative testing (IT) is used to confirm the integrity of gastrointestinal anastomosis. Clinical trials are available in the literature to support the fact that methylene blue can identify the leaks, and can thus help in minimizing the postoperative ratio of clinical leaks after total gastrectomy. In the recent literature, machine learning tools have been used very successfully to investigate the hypothesis of such complex clinical trials, where incomplete data is available. In this article, data obtained from a clinical study, is analyzed using machine learning, to verify whether or not the methylene blue test can accurately identify the leaks and to predict future outcomes. Furthermore, a comparative study based on most robust machine learning solvers is presented in this article to identify the most appropriate machine learning technique(s) for future applications. We have considered the data (over a period starting from Jan 2007 till Dec 2014) based on the total gastrostomies (TG), where methylene blue test was applied. Data was obtained from 198 patients having gastric cancer. Out of 198, 108 cases went through methylene blue test done by a nasojejunal tube while no test was carried out for rest of 90 cases. Intraoperative leakage rate, mortality rate, length of hospitalization and postoperative clinical leakage rate were the measured outcomes. To analyze the data and to predict whether there will be a leak or not, machine learning techniques were applied and the accuracy was compared. The main objective of this research is to predict the clinical leakage after applying methylene blue test on gastric cancer patients. This objective is successfully achieved by implementing six machine learning approaches. Case specific machine learning approaches are discussed to evaluate post clinical leakage rate and radio leakage rate. From our analysis, we have concluded that the prediction of intraoperative leak, post clinical leak and radio leak is possible with the aid of different machine learning techniques. An important conclusion drawn from this study is that a single machine learning technique can not accurately predict different stages of leak, since the accuracy of the technique depends on the specification of clinical data that varies from stage to stage.