Machine Learning Techniques

We have already explored the meaning of Machine Learning and also discussed some types of learning such as supervised learning, unsupervised learning and deep learning. But then, the question is, how do these machines learn since they are thought how to learn and act like humans? The answer is that there are many techniques which are employed in teaching computers easily and those are our focus in this chapter.

However, before I begin to explain each of the Machine Learning techniques listed below, it will be important to refresh our minds on the meaning of the word “technique”. Technique means the way through which a problem is solved.  Let’s say you have a big data and wants to sort them according to similarities, what will you do? You need a technique that will help you to handle such a Herculean task.

Likewise, in Machine Learning, when a computer faces such a task of grouping big data based on similarities, it will need a technique known as “classification” to handle it efficiently.

Some of the main techniques of Machine Learning include; regression, classification, clustering, anomaly detection, market-basket analysis are examined below.

Regression

Regression is one of the statistical processes that enable analysts to estimate the relationships existing among variables. With regression analysis, you can easily understand the changes that occur in the value of a dependent variable if any independent variable changes while others are fixed. Therefore, regression analysis can help you to model and analyze many variables. Now you may wonder what it has to do with Machine Learning (Wikimedia Foundation Inc, 2018).

In Machine Learning, regression algorithm is necessary because of its importance when it comes to prediction & forecasting. When there is adequate data, predicting the outcome of a particular event will be easy, based on the relationship that exists between the variables gotten from it.

Regression algorithm is part of the supervised learning algorithm (Bhatia, 2017). Since it is under supervised learning, the algorithm can use the features of the data in the system to predict an accurate output value. Regression models have many applications such as trend analysis, financial forecasting, marketing, drug response modelling and time series predictions.

Although there are many types of regression, the popular ones in Machine Learning are linear regressions, Lasso regression, multivariate regression and regression trees (Math Works, 1994).

Classification

Another Machine Learning technique is classification. It falls under the supervised machine learning algorithm. A computer program learns to use the data input which programmers feed into it. With this input, it can classify observations easily. Sometimes, the data which the program uses to learn may either be bi-class or multi-class. When it is bi-class, it can be able to identify the sex of a person, whether the person is female or male or to classify emails as either spam emails or non-spam emails.

Some of the classification problems which computers solve are handwriting recognition, document classification, speech recognition, biometric identification etc. In Machine Learning, the following classification algorithms are used: Linear classifiers, support vector machines. Decision trees, boosted trees, random forest, neural networks, nearest neighbor, etc (Sidana, 2017).

Clustering

Clustering is a method of unsupervised Machine Learning. We already know that when learning is unsupervised, computers draw an inference from a set of data without labels. Now with the clustering technique, computer programs can divide data points or a particular population into groups. When the data points are given, the clustering algorithm can help to classify the groups in such a way that data points in the same group will have the same features or similarities, while the data points in different groups will have properties and features that are dissimilar (Nagpal, 2017).

When it comes to data science, clustering helps you to understand available data, more especially when you can see the groups to which the data points fall into with the application of clustering algorithm. Some of the popular clustering algorithms include:

  • K-Means clustering
  • Hierarchical clustering.

K-Means clustering – This is one of the clustering algorithms which you will be taught in an introductory class of Machine Learning and data science. One good thing about this algorithm is that one can understand it very fast and also implement it in code. One of the advantages of K-Means is that it is fast and all that is required of you is to do a few computations of distances that exist between points and different group centres.

However, there are some disadvantages using the K-Means algorithm. The first one is that you must select the number of classes/groups in the data. Also, you may have different results every time you run the algorithm. As such, you may have trouble achieving consistency.

Hierarchical clustering – This type of clustering algorithm comes in two categories; bottom-up or top-down. The bottom-up treats every data point singly from the beginning, and as it continues, it will merge clusters continuously till every cluster becomes one single cluster containing all the data points. Few notable points about hierarchical clustering are that you don’t need to specify how many the clusters are, and you can select which cluster you want to use. Due to the efficiency of the hierarchical clustering algorithm, you can reduce your costs of operations (Seif, 2017).

Someone may wonder why Machine Learning requires the use of clustering. Well, when you group similar entities, it will be easy to profile their attributes based on their groups. At least, you can gain insight into the patterns of the groups. For example, when you group unlabelled data, you can easily identify the existing groups of your customers and apply different marketing strategies to each group.

Anomaly Detection

One important Machine Learning technique is what is known as anomaly detection. This technique is very significant during data mining. It is a technique used to identify events or items that are rare in comparison to the whole data. When there is an anomaly in a data set, it means that there are some items that are not the same with others in the set or that do not follow the expected pattern.

Ordinarily, when there are anomalous items in a data set, they have to be fished out or else they may cause problems when the data is used. Some of the problems caused by these anomalies are medical problems, structural defects, and bank fraud or text errors. Other names for these anomalies are outliers, noise, exceptions, novelties, and deviations (Wikimedia Foundation Inc., 2018).

The importance of using Machine Learning in anomaly detection is to quicken the speed by which these rare items are identified. When companies use Machine Learning algorithms, they can easily and effectively detect and classify the anomalies existing in data sets. The algorithms are capable of learning from a set of data and predict events based on it. Machine Learning that helps in detecting anomalies include some techniques that can help detect and also classify the anomalies according to the large features of the data (Josh, 2017).

Two of the most important Machine Learning techniques for anomaly detection are; Supervised and Unsupervised machine learning for detecting anomalies (Josh, 2017).

Market Basket Analysis

Market Basket analysis is a data analysis technique which is often employed in retailing and marketing. Retailers use the MBA technique to predict or understand their customers purchasing behaviors. The aim of using this analysis is to determine the types of products customers buy together (Megaputer Intelligence Inc, 2002).

You may wonder at the name, but remember when you went to grocery shopping recently. You probably picked a basket and started throwing different items into it. Now, the types of products or items you bought together are what MBA analyzes to understand your buying behavior.

Market Basket Analysis is critical to both companies and retailers. With the information gathered from MBA, a store can rearrange its products to keep those items that customers buy together in one area. Also, merchants who sell over the Web can also use the same information to build a suitable catalog, while it will also guide direct marketers in determining the new products they can offer to their customers (Megaputer Intelligence Inc, 2002).

Market Basket Analysis has two types namely; Predictive MBA and Differential MBA.

  • Predictive Market Basket Analysis: This type of MBA is suitable for classifying events, services and items purchases that occur in a sequential pattern.
  • Differential MBA: This technique aims at removing insignificant results and leaves the analyst with in-depth results. It does not just depend on the buying behaviours of customers in a particular shop. Instead, it compares the information from different stores, weekdays, demographics, yearly seasons, etc.

Apart from analyzing the shopping pattern of customers, there are other areas where MBA is applied.

  1. Analyzing purchases made through credit cards.
  2. Analyzing the patterns of telephone calls.
  3. Identifying medical insurance claims that are fraudulent.
  4. Analyzing purchases of telecom service (Techopedia Inc, 2018).

Time Series Data

One thing that self-driving cars, trading algorithms, transportation networks and smart homes have in common is the need for a unique type of data. This type of data measures the changes that occur in the environment over time, and it is called time series data (TDS).  A time series data shows how a process, system, or behaviour changes over time.

Since time series data records every change that occurs as new, you can measure the change in the system, analyze how it occurred in time past, monitor the present situation and predict the nature of future changes.

We have many kinds of TSDs such as mobile/web App event streams, scientific measurements, DevOps monitoring data, industrial machine data.

The importance of Time Series Data cannot be emphasised as many industries now use it on large datasets because it helps in situations where single data points don’t give the expected results. Also, Time Series helps in forecasting. It influences the results of risk analysis when applied to financial dealings. It also helps to predict machine learning algorithms and meteorology.

Another important use of Time Series Data is in the area of anomaly detection. It makes it easy to understand patterns and extract relevant information, eliminate noise and finally discover anomalies.

The uses of Time Series Data are wide. Internet of Things systems employs it to manage the regular influx of data. Also, DevOps uses Time Series Data to track trends and system health. In recent times, TSDBs have been growing more than other categories of databases. Two main reasons for this continuous growth are scale and usability (Kulkarni, 2017).

Due to the high efficiency and accuracy level of Time Series Data, engineers, tinkerers, scientists and others use it in different applications; some of the cases where TSDBs are applied are:

  • Physical system monitoring: machinery, our bodies, equipment, our homes, the environment, etc.
  • Software systems: containers, applications, virtual machines, services, etc.
  • Financial trading systems
  • Asset tracking applications
  • Creating Eventing Apps
  • Developing business intelligence tools, etc (Kulkarni, 2017).

Neural Networks

The focus of Artificial Intelligence is the development of computer systems so that they perform human-like activities effectively. One of the algorithms that make Artificial Intelligence effective is what is fondly known as Artificial Neural Network. ANN is a deep learning technology that teaches computer systems how to act like humans.  As such, the neural network system is designed based on how the neurons in our brain work. However, even though ANN operates like the brain, the human brain is still superior (Rouse, 2018).

Some of the applications of Artificial Neural Networks include pattern recognition or solving difficult signal processing problems. Examples of these applications are handwriting recognition, weather prediction, speech-to-text transcription, facial recognition and data analysis for oil exploration.

So how does a neural network operate as a human brain? It has many processors that operate in parallel and also arranged in layers. Raw data comes into the first layer and delivers output to the next layer which in turn processes it and sends to the next layer. When it gets to the last layer, it will do the final processing and deliver the system output. One notable thing about neural networks is that they are adaptive. They can continue to adjust themselves from the first training and continue to learn from subsequent runs.

When it comes to how ANNs become skilled at performing their tasks, the answer is simple. The first training involves the input of large data and programming the neural network with an expected output from the input. For instance, if you want to build a system that can recognize the faces of governors, the first training step might be to bring different pictures of governors, doctors, musicians, animals etc. Then as you add any of the input, you will also add a matching identification. This process helps the model to do a better job (Rouse, 2018).

There are many types of neural networks such as feed-forward, recurrent neural networks, and convolutional neural networks. Some of the applications of neural networks include translation, language processing, chatbots, stock market prediction, drug discovery & development, etc. These are specific areas where the application of artificial neural networks is pronounced.

However, other areas include any process which has large data and some strict patterns and rules governing its operations. Therefore, when you are faced with a huge amount of data, and you are unable to understand it within a given time frame, the best option is to mechanize the whole process by using artificial neural networks.

Leave a Reply

Your email address will not be published. Required fields are marked *