In class, we discussed two techniques for dimensionality reduction. What advanta

Question

In class, we discussed two techniques for dimensionality reduction. What advantages do Auto encoders have over Principal Components Analysis (PCA)? We have acquainted ourselves with two broad areas for feature engineering. Elaborate on| the differences between feature extraction and feature selection. Provide a valid reason why feature extraction is an important topic before running a data mining method. In class, we discussed the importance of the order of the data mining method. Elaborate on the differences between running K-means Say you were given an unlabelled data set with regards to their customers on their website. Propose an experimental design to predict the loyalty of an unknown customer. In class, we discussed two approaches for "pipe lining" data mining techniques with PCA and K-means. One was to apply PCA first on the raw data and then apply K-means. Another approach was to apply K-means on the raw data and then PCA. While both approaches are valid, they have very different results. Explain the differences between the two approaches.

Explanation / Answer

1)Autoencoder/Auto Associative neural networks are neural networks that are trained to recall their inputs. Thus the number of inputs is equal to the number of outputs. Autoencoder neural networks have a bottleneck that results from the structure of the hidden nodes. There are less hidden nodes than input nodes. This results in a butterfly structure. The autoencoder network is preferred in recall applications as it can map linear and nonlinear relationships between all of the inputs. The autoencoder structure results in the compression of data into a smaller dimension and then decompressing into the output space. Autoencoders have been used in a number of applications including missing data imputation.

2)FEATURE EXTRACTION :

FEATURE SELECTION

3)

Related Questions

Navigate

• Browse (All)
• Browse I
• Subjects

---

---

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.