**Figure 9.**

*Sparse PCA.*


**Figure 10.** *Randomized PCA.* *Evaluation of Principal Component Analysis Variants to Assess Their Suitability… DOI: http://dx.doi.org/10.5772/intechopen.105418*

#### **Figure 11.** *Incremental PCA.*

## **6.2 Observations**

Based on the results obtained from the variants of PCA for mobile malware detection depicts that for the given CICMalDroid\_2020 dataset is discussed. It is a numerical labelled data that highly supports normal standard PCA technique. It helps to reduce the dimensions of the PCA from 471 features into two sets of PCs (PC1 and PC2). It supports to processing the data model quickly and forms the clusters

#### **Figure 13.**

*Class of malwares in the CICMalDroid\_2020 dataset.*


#### **Table 1.**

*CICMalDroid\_2020 dataset—sample size*

effectively. **Figure 13** shows the group of clusters that discovered the five different malware involved in the CICMalDroid\_2020 dataset based on PC1containing huge information about the dataset of normal PCA.

**Table 1** shows the size of malware samples available under the category of Adware, Banking malware, SMS malware, Riskware, and Benign.

Hence, the other variants of PCA, such as sparse PCA are applicable for sparse data, randomized PCA and incremental PCA are suitable for big data processing and kernel PCA is widely supported by non-linear data modelling. So, depending upon the type of data and their accessibility, the appropriate type of PCA is incorporated into machine learning algorithms specifically for unsupervised learning for dimensionality reduction to develop a suitable predictive model. It also helps to identify the hidden patterns of the data effectively.
