Globally, breast cancer is the most prevalent and leading cause of cancer-related death. Despite the improvement of treatment, the mortality rates can vary by geographical location and lifestyles; in fact, they are population-specific. Patient outcomes would be greatly improved by early detection by means of identifying aberrant germline mutations based on their traits. The proposed research hypothesised to create a tailored neural network (NN) based decision-making model to predict breast cancer with a germline dataset (DS) and assess the effect of oversampling methods on model performance. NN models were trained using a germline DS. To deal with the issue of class imbalance, three methods of oversampling - adaptive synthetic sampling (ADASYN), Synthetic Minority Over-sampling Technique with Edited Nearest Neighbors (SMOTE-ENN), and Borderline SMOTE - were used, producing three balanced DSs along with the imbalanced DS. On each DS, random forest feature selection was used to determine the ten most important features. NN models were further customised, trained, and tested on the imbalanced and balanced DSs. The NN models were highly performing on both the DSs: on balanced DSs, they yielded an accuracy of 98%, precision of 99%, a recall of 97%, and an F1-score of 98%. Recall and F1-score on the unbalanced DS were 99%. The ADASYN and SMOTE-ENN balanced DSs were shown to have perfect discrimination, exhibiting 100% true positive value at zero false positive value by receiver operating characteristic analysis. The NN system proposed together with the effective feature selection and oversampling methods correctly predicts breast cancer on the germline data. The chosen essential features and excellent diagnostic results confirm the model as a possible tool in clinical diagnostic methods for detecting high-risk mutations, which allows early detection and, possibly, increases the outcome of survival.