Comprehensive Model Comparison for Intrusion Detection in UNR-IDD Dataset: Evaluating Naïve Bayes, Decision Tree, Random Forest, K-Neighbors, and LSTM

This paper compares the efficiency of five supervised learning algorithms, Namely Naïve Bayes, Decision Tree, Random Forest, K-Neighbors (KNN), and Long Short-Term Memory (LSTM) on intrusion detection using the UNR-IDD dataset. We analysed the results of the models considering the accuracy, precision, and F1-score. The Decision Tree, Random Forest, and LSTM models were also shown to be the best performers with scores of 1 for accuracy, F1-score, and area under the curve on the testing set. The Naïve Bayes yielded low standard error of 0.038165 but the precise values of precision at 0.773218 and F1-score at 0.872107 depict how the model contributed slightly fewer true positives but more false positives. From such outcomes obtained above, it is evident that Decision Tree, Random Forest and LSTM have high accuracy and appropriateness for this intrusion detection problem, although the accuracies of 100\% are questionable because of the possibilities of over fitting. In terms of classification too, K-Neighbors has very good results and rarely misclassifies patterns. Despite this, Naïve Bayes is not the most suitable method in the present case for this particular dataset. This analysis also demonstrates the specific advantages and disadvantages of each model and gives the understanding of real-world usability of intrusion detection systems.