All the code examples can be found in the Chapter 02 folder in GitHub.

Machine learning dates back to centuries. It was born from the theory that computers can learn without being programmed to perform specific tasks. The iterative aspect of ML is essential as the machines need to adapt themselves to new data always. They need to learn from the historical data, optimize for better computations, and also generalize themselves to provide proper results.

We all are aware of rule-based systems, where we have a set of predefined conditions for a machine to execute and provide the results. How great will it be when machines learn these patterns by themselves, deliver the results, and explain the rules that it discovered; this is ML. It is a broader term used for various methods and algorithms that are used by machines to learn from the data. As a branch of artificial intelligence (AI), the ML algorithms are quite often used to discover...

Let's begin our triple W session with linear regression first.

It is the traditional and most-used regression analysis. It is studied rigorously and used widely for practical purposes. Linear regression is a method for determining the relationship between a dependent variable (y) and one or more independent variables (x). This derived relationship can be used to predict an unexplained y from observed x's. Mathematically, if x is an independent variable (commonly known as the predictor) and y is a dependent variable (also known as the target), the relationship is expressed as follows:

Where m is the slope of line, b is the intercept of the best-fit regression line, and...

Assessing the value of a ML model is a two-phase process. First, the model has to be evaluated for its statistical accuracy, that is, whether the statistical hypotheses are correct, model performance is outstanding, and the performance holds true for other independent datasets. This is accomplished using several model evaluation metrics. Then, a model is evaluated to see if the results are as expected as per business requirement and the stakeholders genuinely get some insights or useful predictions out of it.

A regression model is evaluated based on the following metrics:

• Mean absolute error (MAE): It is the sum of absolute values of prediction error. The prediction error is defined as the difference between predicted and actual values. This metric gives an idea about the magnitude of the error. However, we cannot judge...

Let's start again with the triple W for logistics regression. To reiterate the tripe W method, we first ask the algorithm what it is, followed by where it can be used, and finally by what method we can implement the model.

Logistic regression can be thought of as an extension to linear regression algorithms. It fundamentally works like linear regression, but it is meant for discrete or categorical outcomes.

Logistic regression is applied in the case of discrete target variables such...

Most of the metrics used to assess a classification model are based on the values that we get in the four quadrants of a confusion matrix. Let's begin this section by understanding what it is:

• Confusion matrix: It is the cornerstone of evaluating a classification model (that is, classifier). As the name stands, the matrix is sometimes confusing. Let's try to visualize the confusion matrix as two axes in a graph. The x axis label is prediction, with two values—Positive and Negative. Similarly, the y axis label is actually with the same two values—Positive and Negative, as shown in the following figure. This matrix is a table that contains the information about the count of actual and predicted values by a classifier:

• If we try to deduce information about each quadrant in the matrix:
  • Quadrant...

Decision trees are extensively-used classifiers in the ML world for their transparency on representing the rules that drive a classification/prediction. Let us ask the triple W questions to this algorithm to know more about it.

Decision trees are arranged in a hierarchical tree-like structure and are easy to explain and interpret. They are not susceptive to outliers. The process of creating a decision tree is a recursive partitioning method where it splits the training data into various groups with an objective to find homogeneous pure subgroups, that is, data with only one class.

SVM is a supervised ML algorithm used primarily for classification tasks, however, it can be used for regression problems as well.

SVM is a classifier that works on the principle of separating hyperplanes. Given a training dataset, the algorithms find a hyperplane that maximizes the separation of the classes and uses these partitions for the prediction of a new dataset. The hyperplane is a subspace of one dimension less than its ambient plane. This means the line is a hyperplane for a two-dimensional dataset.

SVM...

Before we build a KNN model for the HR attrition dataset, let us understand KNN's triple W.

KNN is one of the most straightforward algorithms that stores all available data points and predicts new data based on distance similarity measures such as Euclidean distance. It is an algorithm that can make predictions using the training dataset directly. However, it is much more resource intensive as it doesn't have any training phase and requires all data present in memory to predict new instances.

Euclidean distance is calculated as the square root of the sum of the squared differences between two points.

Ensembling models are a robust approach to enhancing the efficiency of the predictive models. It is a well-thought out strategy that is very similar to a power-packed word—TEAM !! Any task done by a team leads to significant accomplishments.

Likewise, in the ML world, an ensemble model is a team of models operating together to enhance the result of their work. Technically, ensemble models comprise of several supervised learning models that are individually trained, and the results are merged in various ways to achieve the final prediction. This result has higher predictive power than the results of any of its constituting learning algorithms independently.

Mostly, there are...

We have created around six classification models on the HR attrition dataset. The following table summarizes the evaluation scores for each model:

The random forest model appears to be a winner among all six models, with a record-breaking 99% accuracy. Now, we need not further improve the random forest model, but check whether it can generalize well to a new dataset and the results are not overfitting the train dataset. One of the methods is to do cross-validation.