Artificial intelligence | June 23, 2021 ©GETBES | 0 Views

1.Linear Regression-

In statistics and machine learning, linear regression is one of the most well-known and well-understood techniques. .

At the price of explain ability, predictive modelling is primarily concerned with decreasing a model's error or creating the most accurate forecasts feasible. We'll take algorithms from a variety of domains, including statistics, and utilize them to achieve these goals.

Linear regression is represented by an equation that defines a line that best matches the connection between the input variables (x) and the output variables (y) by determining precise weightings for the input variables, known as coefficients (B).

**2.Linear Discriminant Analysis-**

The classification algorithm logistic regression has typically been confined to two-class classification issues. The Linear Discriminant Analysis method is the preferred linear classification approach when there are more than two classes.

LDA is represented in a straightforward manner. It is made up of statistical characteristics determined for each class of your data.

Calculating a discriminating value for each class and producing a forecast for the class with the highest value is how predictions are created. Because the methodology implies that the data has a Gaussian distribution (bell curve), it's a good idea to eliminate outliers from your data first. It's a straightforward and effective strategy for solving classification and predictive modelling challenges.

3.Naive Bayes-

For predictive modelling, Naive Bayes is a basic but surprisingly strong algorithm.

The model is made up of two types of probabilities:

1) the likelihood of each class, and

2) the conditional probability for each class given each x value, both of which may be derived directly from your training data. Using Bayes Theorem, the probability model may be used to create predictions for fresh data once it has been calculated. When working with real-valued data, it's typical to assume a Gaussian distribution (bell curve) to make estimating these probabilities easier.

Naive Bayes is called naive because it assumes that each input variable is independent. Despite the fact that this is a strong assumption that is unrealistic for real data, the technique is very effective on a wide range of complex problems.

4. K-Nearest Neighbors-

The KNN algorithm is both simple and powerful. The full training dataset is used as the model representation for KNN. Isn't it simple?

For each new data point, predictions are formed by exploring the whole training set for the K most similar examples (neighbors) and summing the output variable for those K examples. This might be the mean output variable in a regression issue, or the modal (or most common) class value in a classification issue.

The challenge is figuring out how to identify how similar the data instances are. If your characteristics are all on the same scale (for example, all in inches), the easiest method is to utilize the Euclidean distance, which you can compute straight from the differences between each input variable.

KNN can use a lot of memory or space to keep all of the data, but it only calculates (or learns) when a prediction is required, and only when it is needed. To keep your predictions, correct, you may update and curate your training instances over time.

The concept of distance or proximity can break down in very high dimensions (many input variables), which might have a detrimental impact on the algorithm's performance on your challenge. This is known as the dimensionality curse. It recommends that you just use input factors that are most important in predicting the output variable.

5.Learning Vector Quantization-

The fact that you must keep the full training dataset is a disadvantage of K-Nearest Neighbors. The Learning Vector Quantization method (or LVQ for short) is an artificial neural network approach that lets you pick how many training examples to keep and then learns exactly how those examples should appear.

A collection of codebook vectors is used to represent LVQ. These are chosen at random at first, then altered over a number of rounds of the learning process to best summarizes the training dataset. The codebook vectors may be used to create predictions in the same way as K-Nearest Neighbors can. Calculating the distance between each codebook vector and the new data instance yields the most comparable neighbor (best matching codebook vector). The best matched unit's class value (or real value in the case of regression) is then returned as the prediction. If you rescale your data to have the same range, such as between 0 and 1, you'll get the best results.

If you find that KNN performs well on your dataset, consider utilising LVQ to decrease the amount of memory required to store the whole training dataset.

6.Logistic regression-

Logistic regression is another statistical tool that machine learning has taken. It's the approach of choice for binary classification issues (problems with two class values).

Like linear regression, the purpose of logistic regression is to identify the values for the coefficients that weight each input variable. In contrast to linear regression, the output prediction is modified using a non-linear function known as the logistic function.

The logistic function, which resembles a large S, transforms any value into a range of 0 to 1. This is beneficial because we can use a rule to snap values to 0 and 1 (e.g., if less than 0.5, output 1) and forecast a class value using the logistic function's output.

The predictions provided by logistic regression may also be utilized as the likelihood of a particular data instance belonging to class 0 or class 1 due to the way the model is learnt. This might be beneficial in situations when you need to provide more support for a forecast.

When you exclude qualities that are unrelated to the output variable as well as qualities that are highly similar (correlated) to each other, logistic regression works better, much as linear regression. It's an easy-to-learn model that works well with binary classification tasks.

- by Shubham Jha, Editor.