Current SMS spam detection systems typically utilise spam filters, which are often implemented by mobile service providers to identify and automatically block spam messages, preventing them from reaching users’ inboxes. This research presents an enhanced SMS spam detection model. The dataset used in the study is the SMS Spam Collection dataset sourced from Kaggle. Text vectorisation was achieved using bag-of-words, term frequency-inverse document frequency, and Bi-gram (BG) techniques. In this research, a three-step data preprocessing method was employed to improve the model’s performance. This process included: (1) text vectorisation using BG, (2) combining BG data with feature extraction (FE), and (3) addressing the imbalance in the SMS Spam Collection dataset with a class balancer (CB). These three steps resulted in the creation of a dataset referred to as BG-FE-CB data. The model development utilised several well-established algorithms, including naive Bayes, multinomial naive Bayes, logistic regression, support vector machine (SVM), and convolutional neural networks. The models were evaluated using various metrics such as sensitivity, specificity, area under the receiver operating characteristic curve (AUC), the Matthews correlation coefficient (MCC), and accuracy. The experimental results demonstrated that the SVM model using BG-FE-CB data effectively addressed cost-sensitive issues. This proposed model, called BG-FE-CB-SVM, achieved the highest sensitivity (0.95), specificity (1.00), and AUC (0.99) among the models evaluated in this study. It also outperformed previous studies, achieving the highest sensitivity, specificity, AUC, and MCC (0.94).