Today’s world is well advanced and moving towards more sophistication. Due to this, several diseases have occurred, out of which cancer is always a deadly disease as reported in all parts of the world. On top of this, breast cancer is becoming more brutal in human lives, since it has higher mortality in women. Thus, there is a demand always for breast tumor detection among researchers. The work intends to design a robust framework for the classification of breast tumors using Long short-term memory (LSTM) and Boosting models. Before classification, the work makes use of an efficient Chaotic Crow-Search Optimization (ChCSO) algorithm for feature selection. The chaotic maps are integrated with the simple crow-search optimization algorithm to provide a better selection of salient features in the dataset. For evaluation, the work adopted the standard Wisconsin Breast Cancer Diagnosis (WBDC) database. For comparative analysis, the performance of LSTM, AdaBoost, and XGBoost algorithms are compared with and without a feature selection approach. As a result, the LSTM with the ChCSO algorithm provides a superior performance of accuracy of 98.56% comparatively.