The Internet of Things (IoT) has become an inevitable component in the present scenario. At the same time, the introduction of recent high performing algorithms has increased the risk to the security of network communications. Identifying the pattern of attack is a way to address the problem. Once the proper classification is done, such a classification model is of significant reliability. This paper recommends a stacking-enhanced voting ensemble model(SEVEM) designed specifically for IoT-based networks as a strong solution for network intrusion detection. The model combines various base classifiers and meta-learners, combined by soft voting, to improve detection accuracy and generalization both in binary and multiclass classification problems. We conducted in-depth preprocessing with an aggregate feature ranking (AFR) approach incorporating chi-square, mutual information (MI), pearson correlation, and coefficient determination ranking (CDR) for choosing top features, together with synthetic minority over-sampling technique combined with edited nearest neighbors (SMOTEENN) to manage extreme class imbalance. Experimental comparisons on two benchmark datasets, NF-BoT-IoT and NF-ToN-IoT, prove that the presented model exceeds numerous state-of-the-art approaches. The evaluations verify that our model is effective, interpretable, and appropriate for real-time intrusion detection in IoT environments.