Mrmr+ and Cfs+ feature selection algorithms for high-dimensional data

A new population initialization of metaheuristic algorithms based on hybrid fuzzy rough set for high-dimensional gene data feature selection

In the realm of modern medicine and biology, vast amounts of genetic data with high complexity are available. However, dealing with such high-dimensional data poses challenges due to increased processing complexity and size. Identifying critical genes to reduce data dimensionality is essential. The filter-wrapper hybrid method is a commonly used approach in feature selection. Most of these methods employ filters such as MRMR and ReliefF, but the performance of these simple filters is limited. Rough set methods, on the other hand, are a type of filter method that outperforms traditional filters. Simultaneously, many studies have pointed out the crucial importance of good initialization strategies for the performance of the metaheuristic algorithm (a type of wrapper-based method). Combining these two points, this paper proposes a novel filter-wrapper hybrid method for high-dimensional feature selection. To be specific, we utilize the variant of bWOA (binary Whale Optimization Algorithm) based on Hybrid Fuzzy Rough Set to perform attribute reduction, and the reduced attributes are used as prior knowledge to initialize the population. We then employ metaheuristics for further feature selection based on this initialized population. We conducted experiments using five different algorithms on 14 UCI datasets. The experiment results show that after applying the initialization method proposed in this article, the performance of five enhanced algorithms, has shown significant improvement. Particularly, the improved bMFO using our initialization method: fuzzy_bMFO outperformed six currently advanced algorithms, indicating that our initialization method for metaheuristic algorithms is suitable for high-dimensional feature selection tasks.

Poplar's Waterlogging Resistance Modeling and Evaluating: Exploring and Perfecting the Feasibility of Machine Learning Methods in Plant Science

Floods, as one of the most common disasters in the natural environment, have caused huge losses to human life and property. Predicting the flood resistance of poplar can effectively help researchers select seedlings scientifically and resist floods precisely. Using machine learning algorithms, models of poplar's waterlogging tolerance were established and evaluated. First of all, the evaluation indexes of poplar's waterlogging tolerance were analyzed and determined. Then, significance testing, correlation analysis, and three feature selection algorithms (Hierarchical clustering, Lasso, and Stepwise regression) were used to screen photosynthesis, chlorophyll fluorescence, and environmental parameters. Based on this, four machine learning methods, BP neural network regression (BPR), extreme learning machine regression (ELMR), support vector regression (SVR), and random forest regression (RFR) were used to predict the flood resistance of poplar. The results show that random forest regression (RFR) and support vector regression (SVR) have high precision. On the test set, the coefficient of determination (R²) is 0.8351 and 0.6864, the root mean square error (RMSE) is 0.2016 and 0.2780, and the mean absolute error (MAE) is 0.1782 and 0.2031, respectively. Therefore, random forest regression (RFR) and support vector regression (SVR) can be given priority to predict poplar flood resistance.

A Weighted Minimum Redundancy Maximum Relevance Technique for Ransomware Early Detection in Industrial IoT

Ransomware attacks against Industrial Internet of Things (IIoT) have catastrophic consequences not only to the targeted infrastructure, but also the services provided to the public. By encrypting the operational data, the ransomware attacks can disrupt the normal operations, which represents a serious problem for industrial systems. Ransomware employs several avoidance techniques, such as packing, obfuscation, noise insertion, irrelevant and redundant system call injection, to deceive the security measures and make both static and dynamic analysis more difficult. In this paper, a Weighted minimum Redundancy maximum Relevance (WmRmR) technique was proposed for better feature significance estimation in the data captured during the early stages of ransomware attacks. The technique combines an enhanced mRMR (EmRmR) with the Term Frequency-Inverse Document Frequency (TF-IDF) so that it can filter out the runtime noisy behavior based on the weights calculated by the TF-IDF. The proposed technique has the capability to assess whether a feature in the relevant set is important or not. It has low-dimensional complexity and a smaller number of evaluations compared to the original mRmR method. The TF-IDF was used to evaluate the weights of the features generated by the EmRmR algorithm. Then, an inclusive entropy-based refinement method was used to decrease the size of the extracted data by identifying the system calls with strong behavioral indication. After extensive experimentation, the proposed technique has shown to be effective for ransomware early detection with low-complexity and few false-positive rates. To evaluate the proposed technique, we compared it with existing behavioral detection methods.