Breast cancer is one of the major concerns in present day scenario. Detecting breast cancer at early stage increases the chances of survival. The objective of this research is to propose suitable feature selection method to improve the efficiency of breast cancer prediction at early stages to increase the survival rate. In this work, an expert intelligent technique has been proposed named “Expert System for Breast Cancer Prediction (ESBCP)” to detect breast cancer. To validate the results, the proposed system determines accuracy, precision, F-measure, and recall. The proposed model introduced a feature selection technique named-Undiluted Feature Set (UFS) to select the most relevant and promising features. The experimental work was carried out using Python 2.8 version in a Windows environment, taking a dataset on breast cancer from the UCI machine learning repository. There were 699 occurrences in the dataset with nine attributes and two classes. The proposed work utilized a decision tree and a new feature selection technique based on a heuristic search and the Stochastic Hill method. The experimental results were evaluated using the 10-fold Cross-Validation (CV). The experimental findings showed that the suggested model-ESBCP can accurately detect breast cancer at an early stage. As per the result, with simple decision tree the accuracy recorded 93.42 percent whereas ESBCP obtained 94.01 percent. It may seem that the improvement of 0.59 percent is very small, but for a large population even this mere change can have a greater impact. The suggested model ESBCP and the feature selection technique-UFS have a lot of potential in the fields of medical research and bioinformatics in terms of classification capability and predictive power.