Predictive Analytics and Consumer Loyalty

Feature engineering is the process of using domain knowledge of data to create features that make machine learning algorithms work well. The most important reasons to use the selection of the features are:

• It enables machine learning algorithm to train faster.

• It reduces the complexity of the model and makes it easy to interpret.

• It improves the accuracy of the model if the correct subset is selected.

• It reduces overfitting.

Next, we'll discuss several methodologies and techniques that you can use to set your feature space and help your models perform better and more efficiently.

Attribute Selection Algorithms (Features)

Feature selection algorithms discovered and reported in the literature.

The feature selection algorithms are categorized into three categories such as filter model, embedding model (or aggregator) and embedded model according to mathematical models.

Filter model

It depends on the general characteristics of the data and the evaluation of features without involving any learning algorithm. Filter model Algorithms are Relive F, Information Gain. Entropy (H) was used to calculate the homogeneity of a sample.

Information gain is decrease in entropy after dividing the data set by an attribute. Gain index, Chi-Squared, Gain Ratio.

Wrapper model

A predefined learning algorithm is required and its performance is used as a benchmark for evaluation and feature identification.

Embedded model

It chooses a set of features for training and building a model, then test its feature importance depending on the goal of learning model. You can get the feature importance of each feature of your dataset by using the feature importance property of the model. Feature importance gives you a score for each feature of your data, the higher the score more important or relevant is the feature towards your output variable. Feature importance is an inbuilt class that comes with Tree Based Classifiers.

The big data analytics is used technological advances are based on memory usage, data processing and scrutiny of big data include handling high volume data with decreasing cost of storage and CPU power, the effective usage of storage management for flexible computation and storage and Distributed computing systems through flexible parallel processing with the development of new frameworks such as Hadoop. And also big data frameworks such as the Hadoop ecosystem, No SQL databases efficiently handle complex queries, analytics and finally extract, transform and load (ETL) which is complex in conventional data warehouses. These technological changes have shaped a number of improvements from conventional analytics and big data analytics.

The feature selection aims to select a feature subset that has the lowest dimension and also retains classification accuracy so as to optimize the specific evaluation criteria. feature selection methods can be filter and wrapper. Filter methods make a selection on the basis of the characteristics of the dataset itself. The selection may be fast but it leads to poor performance. Wrapper methods takes a subsequent classification algorithm as the evaluation index, it achieves high classification accuracy but results in very low efficiency for a large amount of computation. Backward feature elimination and Forward feature construction have been used. Feature selection methods such as backward elimination, Forward selection.

Backward elimination: In contrast to the forward selection strategy, the backward elimination strategy starts with the complete attribute set as initial subset and iteratively (and also heuristically) removes attributes from that subset, until no performance gain can be achieved by removing another attribute.

Forward selection initially uses only attribute subsets which exactly one attributes. Then additional attributes are added heuristically until there is no more performance gain by adding an attribute. Both of the methods are handled carefully Hadoop ecosystem.