Software development has several stages to be followed by developers, starting from planning to deployment and maintenance. From those stages, the design phase, which serves as a blueprint of the system to be built, needs concentrated and expert work since it is a creative and difficult stage that determines the behavior of the software. Hence, to have high-quality software, we need to have good architecture. Selecting an architectural pattern for software depends on features like Quality Attributes (QA), functional requirements, and constraints. Due to this architects face difficulty in the selection of the right pattern for the system, and we have used tactics as a selection feature since tactics are building blocks of architectural patterns. In this study, we have built a model for the selection of architectural patterns from tactics. To build the proposed model, we have used an experimental research approach. For the experiment, we have prepared six hundred requirement text which was labeled by experts to the corresponding tactics. This dataset used for built tactic classification model. Then we have prepared dataset which shows the influence of tactics on different architectural patterns form different literatures and we used this dataset for recommendation of architectural pattern. The proposed model used Natural Language Processing (NLP) concepts for pre-processing of the datasets. We used Term Frequency-Inverse Document Frequency (TF-IDF) and word2vec vectorization after preprocessing to convert the textual data format to vectorized form. Furthermore, for classification, we used three different machine learning algorithms. To reach out high performance, we ran six different combinations of experiments. These combinations were TF-IDF with Support Vector Machine (SVM), TF-IDF with Naïve Bayes (NB), TFI-IDF with decision tree, pre-trained word2vec with SVM, pre-trained word2vec with NB, and pre-trained word2vec with decision tree. And obtained the accuracy of 94%, 88%, 79%, 57%, 47%, 53; respectively. We selected TF-IDF with SVM combination since it performed better than the other. Finally, we used a custom-built algorithm for the recommendation of an architectural pattern based on the tactics.