Machine learning models are greatly influenced by the dimension of the dataset and the amount of features, as redundant and irrelevant characteristics can drastically affect performance. To overcome this issue, Improved Mutual Information (MI) based feature selection (IMIFS) with Multi-Label (ML)-KNN classification approach for multi-class intrusion detection network is proposed in this research. IMIFS is a feature selection technique that uses Mutual information method and the Feature Selection Method. The local Area Network (LAN) environments were used to test the proposed improved ML-KNN approach for detecting TCP port scanning attacks. The experimental results from the Attack dataset demonstrate that the suggested approach using Multi-label classification utilized on kernel-space performance achieved F1score of 0.979 & Area under Curve (AuC) of 0.99 percent.