Artificial Intelligence (AI) has grown drastically and becomes more popular in all fields (Vernon et al.,2007) in the 21st century. AI makes an era of computer science, automation, mathematical logic, linguistics etc. The main aspects of AI are to extend the views and augment the capacity and efficiency of mankind in tasks of remarking nature and governing the society through intelligent learning concept (Bench-Capon & Dunne, 2007). Since the 1980s, It has been sub divided into many major areas like computer vision, robotics, game theory, machine learning, etc (Marr, 1977; Nilsson, 2014).

Machine learning makes the system to learn without direct programming. Since, we have training data, it is possible to acquire more accurate models related to that data. It enables models to be trained on data sets before deploying. Some models of machine learning are online, similar to computers (Louridas & Ebert, 2016). This process helps humans, and human supervision is enabled. After the system has been trained, it is ready for real time data. Training session results in more precision and accuracy. The various types of Machine Learning (ML) are Supervised Learning, Unsupervised Learning, and Reinforcement Learning. Machine learning requires the accurate data for learning process. With the help of big data, now it is possible to acquire most accurate information for large enterprises. In addition, robustness and information rate enabled the speeding process, and helps in managing large chunks of data. The hidden patterns and anomalies buried in the data can help or hurt the conclusions. ML requires a planning process for understanding the business problem to be solved and collect the right data sources. However, the reality is that processes change. Therefore, machine learning can make the creation of applications much more dynamic and effective. Supervised Learning requires a pre-set of data and it needs to be further classified for understanding (Kotsiantis, 2007). It is allowed to find errors in the data that can be used for analytics process. This data has labelled features that define the meaning of data. Supervised training models have broad applicability to a variety of business problems, including fraud detection, recommendation solution, and speech recognition or risk analysis. Unsupervised Learning is always handled when the problem requires a large amount of information that is unlabeled. This is applied in social media. It can determine the outcome when there is a massive amount of data. This algorithm can help business understand the large volumes of new unlabeled data. The supervised learning conducts an iterative process for analyzing data without human intervention and unsupervised learning is used with email spam detecting technology, which is the main difference between supervised and unsupervised learning (Zhao & Liu, 2007). The method can determine the outcomes more quickly than other supervised learning. Reinforcement Learning is a similar type of learning model. It requires accurate and precise feedback by analyzing data, so that the user is supervised for the best outcome. It is a special type of learning because it does not deal with sampled data. It learns through performing trial and error method. The reinforcement is applied in robotics, game playing, self- driving cars, etc.

Deep Learning is one of the method of machine learning, which actually deals with the neural networks in the respective layers to learn the data in a recursive manner. It is especially useful to learn patterns from unstructured data. It is a repetitive neural network, which is especially designed to replace the human brain, so that computers can be trained to deal with poorly defined abstractions and problems. It is divided into 3 layers; Input, output layer, and many hidden Layer. Data is enabled through the input layer. Then the data is modified in the hidden layer, and the output layers based on the weights applied to these nodes. It is a machine learning technique that utilizes both supervised and unsupervised learning through analyzing the neural network. Deep learning comes under Machine Learning (Deng & Yu, 2014). Neural networks along with deep learning are often utilized in image recognition, speech, and computer enabled vision applications.

This article describes the various techniques that can be used for each of these three types and discusses the main merits and demerits of each technique with reference to theoretical and factual studies. Further practices are encouraged in applications of machine learning. Figure 1 represents the types of Machine Learning.

Figure 1. Types of Machine Learning

2.1 Supervised Learning

Supervised learning is depends on the data source that trains it for a while. The techniques utilized are the feed forward model. The feed forward model has three important characteristics. The hidden layers help in solving complex problems with the help of neural networks. The problem with neural network is its non-linear characteristics. The inter connection model of the network exhibits a high degree of connectivity. These non-linear characteristics can be overcome by solving difficult problems while establishing these features. We people easily do mistakes while analyzing data mining is also closely related to it. This evolution of smart and Nano technology has created curiosity in finding hidden patterns in data to derive value. In classification, learning is carried out through memory and error detection. Error- Correction Learning, is used with supervised learning, which is a technique for comparing the system output with the desired output value, and using that error to direct the training. Error correction learning algorithms attempt to minimize the error signal at each training iteration. In the most direct route, the error values enable us to directly adjust the tap weights, using an algorithm such as the back propagation algorithm (Sathya & Abraham, 2013). Memory based learning rule decomposes the input space either statically or dynamically into sub regions for the purpose of storing and retrieving functional information. Advantages are in supervised learning, and we have simple process. In the case of unsupervised learning, it is difficult to track how the machine learns. It is easy to find exactly how many classes are there before giving the data for training (Pythonista Planet, 2019). Disadvantages involve complex problems. It cannot differentiate the data on their own.

2.2 Unsupervised Learning

It is the ability to learn data without previous training sessions. Sometime it is advantageous to have a deficient direction, as it lets the algorithm to look for error patterns that was not previously considered. It is intrinsically more difficult than supervised learning because there is no gold standard (like an outcome variable) and no single objective (like test set accuracy). The main objective is to find similarities and then cluster them together (Patrick, n.d.). Hierarchical Clustering comes under the unsupervised learning that does not requires differentiating the groups or clusters. It is an ultimate algorithm, which we have for clustering. Agglomerative contains n clusters, resulting in sub clusters. Steps in agglomerative clustering are as follows: Data points are assigned to each singleton and then recursively iterating the above steps. Calculate the Euclidian distance clusters that it must have a wide dynamic range and flexibility. Usually, small clusters aid for the process. Divisive clustering is closely related to agglomerative clustering but differs in the way they perform, initially which consists of a single cluster and repeat it for successive steps. Steps in divisive clustering are as follows: Initially consider a singleton cluster. To obtain the Euclidian distance, divide the cluster by small clusters. We can access different types of data using Euclidian distance. Computational difficulties arise while splitting the clusters. The results vary based on the distance metrics used.

2.3 Reinforcement Learning

Reinforcement learning algorithm acquires the dataset from the environment in a recursive manner. It explores until it is exposed to maximum states (Sutton & Barto, 2014; 2015). One of the challenges it faces is balancing both exploration and exploitation. There are two types of Reinforcement. They are positive and Negative Reinforecement. Positive Reinforcement has a positive impact on an action. Advantages of reinforcement learning. Ability to retain the change for a long period of time. Disadvantages of reinforcement learning: Excess of data can vanish the result. Negative Reinforcement has a negative impact on an action.

For small representation, machine learning was deployed using Google Colab. Google Colab used to perform machine and deep learning program without much of software's to be installed. The created .json file can be moved to Arduino or RP later on. Figure 2 shows the screen i shot of Google Colab. Here the dataset of doctor appointment is used to predict if the patient will attend the appointment schedule or not based on time duration between scheduled date and appointment fixed date. It also depends on disease history. In Figure 3 the feature correlation is portrayed. Flow of ML is represented in Figure 4.

Figure 2. Screenshot of Google Colab

Figure 3. Feature Correlation

Figure 4. Flow of Machine Learning Algorithm

Figure 3 represents the feature correlation. It is necessary to know about dependencies of variables. It is highly needed to design machine learning models depending upon the requirement of the situation.

Figure 4 describes the data acquisition and prediction of the outcome, the model is then trained and tested and finally implementing in real time.

In this paper, the three main types of Machine Learning is discussed. Combinations of Machine learning, information theory and some mathematical concepts help human to reach a higher level in technology The final analysis is that machine learning has become an inevitable part of technology. This literature survey brings out the need for machine learning and their respective implementations. In comparison, If the training requires known data, it is supervised learning and if the target is not specified, it falls under unsupervised Learning. For large complex problems, reinforcement techniques can be used. Machine learning finds the best solution for a problematic situation.