There are many destructive diseases in the world which cause rapid death by taking time to affect such as cancer and diabetes. They take a lot of time to spread, thus they are curable or somewhat scalable to a great extent if they are diagnosed soon after introduced into the human body. Research shows that almost all type of cancer can be cured if they are detected in the early stage. It is also true for diabetes as they can be controlled if they are detected at the right time. So, a prediction technique that takes help from the computer and processes data from affected user to detect possible contamination can be a great tool for assisting both the doctors and patients with these diseases. A challenge in the process is that the detection accuracy has to be acceptable in order to make the system a reliable one. In this study, we have analyzed medical data using several classification algorithms in order to optimize classifier performance for cancer and diabetes prediction.