Dimensionality reduction is one basic and critical technology for data mining, especially in current “big data” era. It is the process of reducing the number of features (or dimensions) in a dataset while retaining as much information as possible. This can be done for a variety of reasons, such as to reduce the complexity of a model, to improve the performance of a learning algorithm, or to make it easier to visualize the data. Motivated from aforementioned reasons, this paper proposed a new feature reduction approach which reduce and weight the most important features from of universal features set to fit the big data analytics on IoT based cybersecurity systems. The minimal number of features are chosen by using feature selection methods (ANOVA, Variance Threshold, Information Gain, Chi Square) which performed with two files from IoT-23 dataset. According to the approach, we divided the universal features into several subgroups, and evaluated the performance of ML Algorithms (NB, KNN, RF, LR). Extensive experiments are performed with the CICIDS2017 dataset to validate the effectiveness of the proposed approach. As a result, the Random Forest algorithm was the best in terms of performance, as the lowest value of all metrics (Accuracy, Precision, Recall, F1-Score) we obtained was 95%, except for the case in which we used features that we assumed were the least important feature subset. The proposed approach reduced the number of features to only two features and achieved high results.