Data mining. Textbook

© Vadim Shmal, 2022

© Pavel Minakov, 2022

© Sergey Pavlov, 2022

ISBN 978-5-0059-4479-5

Created with Ridero smart publishing system

Data mining

Data mining is the process of extracting and discovering patterns in large datasets using methods at the interface of machine learning, statistics, and database systems, especially databases containing large numerical values. This includes searching large amounts of information for statistically significant patterns using complex mathematical algorithms. Collected variables include the value of the input data, the confidence level and frequency of the hypothesis, and the probability of finding a random sample. It also includes optimizing the parameters to get the best pattern or result, adjusting the input based on some facts to improve the final result. These parameters include parameters for statistical means such as sample sizes, as well as statistical measures such as error rate and statistical significance.

The ideal scenario for data mining is that the parameters are in order, which provides the best statistical results with the most likely success values. In this ideal scenario, data mining takes place within a closed mathematical system that collects all inputs to the system and produces the most likely outcome. In fact, the ideal scenario is rarely found in real systems. For example, in real life this does not happen when engineering estimates for a real design project are received. Instead, many factors are used to calculate the best measure of success, such as project parameters and the current difficulty of bringing the project to the project specifications, and these parameters are constantly changing as the project progresses. While they may be useful in certain situations, such as the development of specific products, their values should be subject to constant re-evaluation depending on the current conditions of the project. In fact, the best data analysis happens in a complex mathematical structure of problems with many variables and many constraints, and not in a closed mathematical system with only a few variables and a closed mathematical structure.

Data is often collected from many different sources and several different directions. Each type of data is analyzed and all of that output is analyzed to get an estimate of how each piece of data may or may not be involved in the final result. Such analysis is often referred to as the analysis process or data analysis. Data analysis also includes identifying other important information about the database that may or may not have a direct impact on the results. Often, they are also generated from different sources.

Data is usually collected from many different sources and many statistical methods are applied to obtain the best statistical results. The results of these methods are often referred to as statistical properties or parameters, and often define mathematical formulas that are intended for the results of each mathematical model. Mathematical formulas are often the most important aspects of the data analysis process and are usually structured using mathematical formulas known as algorithms. Some mathematical algorithms are based on some theoretical approach or model. Other mathematical algorithms use logic and logical proofs as mathematical tools to understand data. Other mathematical algorithms often use computational procedures such as mathematical modeling and mathematical tools to understand a particular problem or data. While such computational procedures may be necessary to complete a mathematical model of the data, such mathematical algorithms may have other mathematical tools that may be more appropriate for the real world. Although these mathematical models are often very complex, it is often easier to develop a mathematical algorithm and model from a mathematical model than to develop a mathematical algorithm and model from an actual data analysis process.

In reality, there are usually a number of mathematical models that provide a more complete understanding of the situation and data than any one mathematical model or mathematical algorithm. The data is then analyzed and a mathematical model of the data is often used to derive a specific parameter value. This parameter value is usually determined by numerical calculations. If a parameter does not have a direct relationship with the result of the final analysis, the parameter is sometimes calculated indirectly using a statistical procedure that yields a parameter that has a direct correlation with the result of the data analysis. If a parameter has a direct correlation with the result of the data analysis, this parameter is often used directly to obtain the final result of the analysis. If the parameter is not directly related to the result of the analysis, the parameter is often obtained indirectly using a mathematical algorithm or model. For example, if data analysis can be described by a mathematical model, then a parameter can be obtained indirectly using a mathematical algorithm or model. It is usually easier to get the parameter directly or indirectly using a mathematical algorithm or model.

By collecting and analyzing many different kinds of data, and performing mathematical analysis on the data, the data can be analyzed and statistics and other statistical tools can be used to produce results. In many cases, the use of numerical calculations to obtain real data can be very effective. However, this process usually requires real-world testing before data analysis.

Agent mining

Agent -based mining is an interdisciplinary field that combines multi-agent systems with data mining and machine learning to solve business problems and solve problems in science.

Agents can be described as decentralized computing systems that have both computing and communication capabilities. Agents are modeled based on data processing and information gathering algorithms such as «agent problem» which is a machine learning technique that tries to find solutions to business problems without any data center.

Agents are like distributed computers where users share computing resources with each other. This allows agents to exchange payloads and process data in parallel, effectively speeding up processing and allowing agents to complete their tasks faster.

A common use of agents is data processing and communication, such as the task of searching and analyzing large amounts of data from multiple sources for specific patterns. Agents are especially efficient because they don’t have a centralized server to keep track of their activities.

Currently, there are two technologies in this area that provide the same functionality as agents, but only one of them is widely used: distributed computing, which is CPU-based and often uses centralized servers to store information; and local computing, which is typically based on local devices such as a laptop or mobile phone, with users sharing information with each other.

Anomaly detection

In data analysis, anomaly detection (also outlier detection) is the identification of rare elements, events, or observations that are suspicious because they differ significantly from most of the data. One application of anomaly detection is in security or business intelligence as a way to determine the unique conditions of a normal or observable distribution. Anomalous distributions differ from the mean in three ways. First, they can be correlated with previous values; second, there is a constant rate of change (otherwise they are an outlier); and third, they have zero mean. The regular distribution is the normal distribution. Anomalies in the data can be detected by measuring the mean and dividing by the value of the mean. Because there is no theoretical upper limit on the number of occurrences in a dataset, these multiples are counted and represent items that have deviations from the mean, although they do not necessarily represent a true anomaly.

Data Anomaly Similarities

The concept of anomaly can be described as a data value that differs significantly from the mean distribution. But the description of anomalies is also quite general. Any number of outliers can occur in a dataset if there is a difference between observed relationships or proportions. This concept is best known for observing relationships. They are averaged to obtain a distribution. The similarity of the observed ratio or proportion is much less than the anomaly. Anomalies are not necessarily rare. Even when the observations are more similar than the expected values, the observed distribution is not the typical or expected distribution (outliers). However, there is also a natural distribution of possible values that observations can fit into. Anomalies are easy to spot by looking at the statistical distribution of the observed data.

In the second scenario, there is no known distribution, so it is impossible to conclude that the observations are typical of any distribution. However, there may be an available distribution that predicts the distribution of observations in this case.

In the third scenario, there are enough different data points to use the resulting distribution to predict the observed data. This is possible when using data that is not very normal or has varying degrees of deviation from the observed distribution. In this case, there is an average or expected value. A prediction is a distribution that will describe data that is not typical of the data, although they are not necessarily anomalies. This is especially true for irregular datasets (also known as outliers).

Anomalies are not limited to natural observations. In fact, most data in the business, social, mathematical, or scientific fields sometimes has unusual values or distributions. To aid decision making in these situations, patterns can be identified relating to different data values, relationships, proportions, or differences from a normal distribution. These patterns or anomalies are deviations of some theoretical significance. However, the deviation value is usually so small that most people don’t notice it. It can be called outlier, anomaly, or difference, with either term referring to both the observed data and the possible underlying probability distribution that generates the data.

Assessing data anomalies problem

Now that we know a little about data anomalies, let’s look at how to interpret the data and assess the possibility of an anomaly. It is useful to consider anomalies on the assumption that data is generated by relatively simple and predictable processes. Therefore, if the data were generated by a specific process with a known probability distribution, then we could confidently identify the anomaly and observe the deviation of the data.

It is unlikely that all anomalies are associated with a probability distribution, since it is unlikely that some anomalies are associated. However, if there are any anomalies associated with the probability distribution, then this would be evidence that the data is indeed generated by processes or processes that are likely to be predictable.

In these circumstances, the anomaly is indicative of the likelihood of data processing. It is unlikely that a pattern of deviations or outliers in the data is a random deviation of the underlying probability distribution. This suggests that the deviation is associated with a specific, random process. Under this assumption, anomalies can be thought of as anomalies in the data generated by the process. However, the anomaly is not necessarily related to the data processing process.

Understanding Data Anomaly

In the context of evaluating data anomalies, it is important to understand the probability distribution and its probability. It is also important to know whether the probability is approximately distributed or not. If it is approximately distributed, then the probability is likely to be approximately equal to the true probability. If it is not approximately distributed, then there is a possibility that the probability of the deviation may be slightly greater than the true probability. This allows anomalies with larger deviations to be interpreted as larger anomalies. The probability of data anomaly can be assessed using any measure of probability, such as sample probability, likelihood, or confidence intervals. Even if the anomaly is not associated with a specific process, it is still possible to estimate the probability of a deviation.

These probabilities must be compared with the natural distribution. If the probability is much greater than the natural probability, then there is a possibility that the deviation is not of the same magnitude. However, it is unlikely that the deviation is much greater than the natural probability, since the probability is very small. Therefore, this does not indicate an actual deviation from the probability distribution.

Revealing the Data Anomalies Significance

In the context of evaluating data anomalies, it is useful to identify the relevant circumstances. For example, if there is an anomaly in the number of delayed flights, it may happen that the deviation is quite small. If many flights are delayed, it is more likely that the number of delays is very close to the natural probability. If there are several flights that are delayed, it is unlikely that the deviation is much greater than the natural probability. Therefore, this will not indicate a significantly higher deviation. This suggests that the data anomaly is not a big deal.

If the percentage deviation from the normal distribution is significantly higher, then there is a possibility that data anomalies are process related, as is the case with this anomaly. This is additional evidence that the data anomaly is a deviation from a normal distribution.

After analyzing the significance of the anomaly, it is important to find out what the cause of the anomaly is. Is it related to the process that generated the data, or is it unrelated? Did the data anomaly arise in response to an external influence, or did it originate internally? This information is useful in determining what the prospects for obtaining more information about the process are.

The reason is that not all deviations are related to process variability and affect the process in different ways. In the absence of a clear process, determining the impact of a data anomaly can be challenging.

Analysis of the importance of data anomalies

In the absence of deviation from the probability distribution evidence, data anomalies are often ignored. This makes it possible to identify data anomalies that are of great importance. In such a situation, it is useful to calculate the probability of deviation. If the probability is small enough, then the anomaly can be neglected. If the probability is much higher than the natural probability, then it may provide enough information to conclude that the process is large and the potential impact of the anomaly is significant. The most reasonable assumption is that data anomalies occur frequently.

Conclusion

In the context of assessing data accuracy, it is important to identify and analyze the amount of data anomalies. When the number of data anomalies is relatively small, it is unlikely that the deviation is significant and the impact of the anomaly is small. In this situation, data anomalies can be ignored, but when the number of data anomalies is high, it is likely that the data anomalies are associated with a process that can be understood and evaluated. In this case, the problem is how to evaluate the impact of the data anomaly on the process. The quality of the data, the frequency of the data, and the speed at which the data is generated are factors that determine how to assess the impact of an anomaly.

Analyzing data anomalies is critical to learning about processes and improving their performance. It provides information about the nature of the process. This information can be used in evaluating the impact of the deviation, evaluating the risks and benefits of applying process adjustments. After all, data anomalies are important because they give insight into processes.

The ongoing process of evaluating the impact of data anomalies provides valuable insights. This information provides useful information about the process and provides decision makers with information that can be used to improve the effectiveness of the process.

This approach makes it possible to create anomalies in the data, which makes it possible to evaluate the impact of the anomaly. The goal is to gain insight into processes and improve their performance. In such a scenario, the approach gives a clear idea of the type of process change that can be made and the impact of the deviation. This can be useful information that can be used to identify process anomalies that can be assessed to assess the effect of deviation. The process of identifying process anomalies is very important to provide valuable data for assessing potential anomalies in process performance.

Anomaly analysis is a process that estimates the frequency of outliers in the data and compares it to the background frequency. The criterion for evaluating the frequency of data deviation is the greater number of data deviations, and not the natural occurrence of data anomalies. In this case, the frequency is measured by comparing the number of data deviations with the background of the occurrence of data deviations.

This provides information on how much data deviation is caused by the process over time and the frequency of deviation. It can also provide a link to the main rejection process. This information can be used to understand the root cause of the deviation. A higher data rejection rate provides valuable insight into the rejection process. In such a situation, the risk of deviation is likely to be detected and necessary process changes can be assessed.

Many studies are conducted on the analysis of data anomalies to identify factors that contribute to the occurrence of data anomalies. Some of these factors relate to processes that require frequent process changes. Some of these factors can be used to identify processes that may be abnormal. Many parameters can be found in systems providing process performance.

Association Rule Learning

Association rule learning is a rule-based machine learning technique for discovering interesting relationships between variables in large sample databases. This technique is inspired by the auditory system, where we learn the association rules of an auditory stimulus and that stimulus alone.

Sometimes when working with a dataset, we are not sure if the rows in the dataset are relevant to the training task, and if so, which ones. We may want to skip those rows in the dataset that don’t matter. Therefore, associations are usually determined by non-intuitive criteria, such as the order in which these variables appear in a sequence of examples, or duplicate values in these data rows.

This problematic aspect of learning association rules can be eliminated in the form of an anomaly detection algorithm. These algorithms attempt to detect non-standard patterns in large datasets that may represent unusual relationships between data features. These anomalies are often detected by pattern recognition algorithms, which are also part of statistical inference algorithms. For example, the study of naive Bayes rules can detect anomalies in the study of association rules based on a visual inspection of the presented examples.

In a large dataset, a feature space can represent an area of an image as a set of numbers, in which each image pixel has a certain number of pixels. The characteristics of an image can be represented as a vector, and we can place this vector in the feature space. If the attribute space is not empty, the attribute will be the number of pixels in the image that belong to a particular color.

Clustering

Clustering is the task of discovering groups and structures in data that are «similar» to some extent, not by using known structures in the data, but by learning from what is already there.

In particular, clustering is used in such a way that new data points are only added to existing clusters, without changing their shape to fit the new data. In other words, clusters are formed before data is collected, rather than fixed after all data is collected.

Given a set of parameters for data that is (mostly) variable, and their «collinearity», clustering can be thought of as a hierarchical algorithm for finding clusters of data points that satisfy a set of criteria. Parameters can be grouped into one of two categories: parameter values that define the spatial arrangement of clusters, and parameter values that define relationships between clusters.

Given a set of parameters for a dataset, clustering can be thought of as discovering those clusters. What parameters do we use for this? The implicit clustering method, which finds the nearest clusters (or, in some versions, clusters more similar to each other) with the least computational cost, is probably the simplest and most commonly used method for doing this. In clustering, we aim to keep the clusters as closely related to each other as possible – whether we do this by taking more measurements or by using only a certain technique to collect data.

But what is the difference between clustering and splitting data into one or more datasets?

The methods of implicit clustering and managed clustering are actually very similar. The only difference is that we use different parameters to determine in which direction we should split the data. Take as an example a set of points on a sphere that define an interconnected network. Both methods aim to keep the network as close as possible to the network defined by the two nearest points. This is because we don’t care if we are very far from one or the other. So, using the implicit clustering algorithm (cluster distance), we will divide the sphere into two parts that define very different networks: one will be the network defined by the two closest points, and the other will be the network defined by the two farthest points. The result is two completely separate networks. But this is not a good approach, because the further we move away from the two closest points, the smaller the distance between the points, the more difficult it will be to find connections between them – since there is a limited number of points that are connected by a small distance.

On the other hand, the method of controlled clustering (cluster distance) would require us to measure the length between each pair of points, and then perform calculations that make the networks closest to each other the smallest distance possible. The result is likely to be two separate networks that are close to each other but not exactly the same. Since we need two networks to be similar to each other in order to detect a relationship, it is likely that this method will not work – instead, the two clusters will be completely different.

The difference between these two methods comes down to how we define a «cluster». The point is that in the first method (cluster distance) we define a cluster as a set of points belonging to a network similar to a network defined by two nearest points. By this definition, networks will always be connected (they will be the same distance apart) no matter how many points we include in the definition. But in the second method (clustering control), we define clusters as pairs of points that are the same distance from all other points in the network. This definition can make finding connected points very difficult because it requires us to find every point that is similar to other points in the network. However, this is an understandable compromise. By focusing on finding clusters with the same distance from each other, we are likely to get more useful data, because if we find connections between them, we can use this information to find the relationship between them. This means that we have more opportunities to find connections, which will make it easier to identify relationships. By defining clusters using distance measurements, we ensure that we can find a relationship between two points, even if there is no way to directly measure the distance between them. But this often results in very few connections in the data.