1C mechanisms for data analysis and forecasting. Data Analysis and Forecasting
Data analysis and forecasting engine- this is one of the mechanisms for generating economic and analytical reporting. It provides users (economists, analysts, etc.) with the opportunity to search for non-obvious patterns in the data accumulated in the information base. This mechanism allows:
- search for patterns in the source data of the information base;
- manage the parameters of the analysis performed both programmatically and interactively;
- provide programmatic access to the analysis result;
- automatically display the analysis result in a spreadsheet document;
- create forecast models that allow you to automatically predict subsequent events or the values of certain characteristics of new objects.
The data analysis mechanism is a set of built-in language objects that interact with each other, which allows the developer to use its components in any combination in any application solution. Built-in objects make it easy to organize interactive configuration of analysis parameters by the user, and also allow you to display the analysis result in a form convenient for display in a spreadsheet document.
The mechanism allows you to work both with data received from the information base and with data received from an external source, pre-loaded into a table of values or a spreadsheet document:
By applying one of the types of analysis to the source data, you can obtain the result of the analysis. The result of the analysis represents a certain model of data behavior. The result of the analysis can be displayed in the final document, or saved for future use.
The further use of the analysis result is that on its basis a forecast model can be created that allows one to predict the behavior of new data in accordance with the existing model.
For example, you can analyze which products are purchased together (in one invoice) and save this analysis result in the database. In the future, when creating the next invoice based on the saved analysis result, you can build a forecast model, feed it “input” with new data contained in this invoice, and “output” receive a forecast - a list of goods that the counterparty B.S. Petrov. He will also most likely acquire them if they are offered to him:
The data analysis and forecasting engine implements several types of data analysis:
Implemented analysis types
general Statistics
It is a mechanism for collecting information about the data in the sample being studied. This type of analysis is intended for preliminary investigation of the data source being analyzed.
The analysis shows a number of characteristics of continuous and discrete fields. Continuous fields contain types such as Number, date. For other types, discrete fields are used. When outputting a report to a spreadsheet document, pie charts are filled in to display the composition of the fields.
Search for associations
This type of analysis searches for frequently occurring groups of objects or characteristic values together, and also searches for association rules. Association search can be used, for example, to determine frequently purchased goods or services together:
This type of analysis can work with hierarchical data, which allows, for example, to find rules not only for specific products, but also for their groups. An important feature of this type of analysis is the ability to work both with an object data source, in which each column contains some characteristic of the object, and with an event source, where the characteristics of the object are located in one column.
To make the result easier to perceive, a mechanism for cutting off redundant rules is provided.
Sequence Search
The sequence search type of analysis allows you to identify sequential chains of events in a data source. For example, this could be a chain of goods or services that customers often purchase sequentially:
This type of analysis allows for hierarchical searches, which makes it possible to track not only the sequences of specific events, but also the sequences of parent groups.
A set of analysis parameters allows a specialist to limit the time distances between elements of the searched sequences, as well as adjust the accuracy of the results obtained.
Cluster analysis
Cluster analysis allows you to divide the original set of objects under study into groups of objects, so that each object is more similar to objects from its group than to objects from other groups. By further analyzing the resulting groups, called clusters, you can determine how this or that group is characterized and decide on methods for working with objects of various groups. For example, using cluster analysis, you can divide the clients with whom a company works into groups in order to apply different strategies when working with them:
Using the parameters of cluster analysis, the analyst can configure the algorithm by which the partitioning will be carried out, and can also dynamically change the composition of the characteristics taken into account in the analysis and configure weighting coefficients for them.
The result of clustering can be displayed in a dendrogram - a special object designed to display sequential relationships between objects.
Decision tree
The decision tree type of analysis allows you to build a hierarchical structure of classifying rules, presented in the form of a tree.
To build a decision tree, you need to select a target attribute on which the classifier will be built and a number of input attributes that will be used to create rules. The target attribute may contain, for example, information about whether the client switched to another service provider, whether the transaction was successful, whether the work was done well, etc. Input attributes, for example, can be the employee’s age, his work experience, the client’s financial condition, the number of employees in the company, etc.
The result of the analysis is presented in the form of a tree, each node of which contains a certain condition. To decide which class a new object should be assigned to, it is necessary, by answering questions at the nodes, to go through the chain from the root to the leaf of the tree, moving to the child nodes in the case of an affirmative answer and to the neighboring node in the case of a negative answer.
A set of analysis parameters allows you to adjust the accuracy of the resulting tree:
Forecast models
Forecast models created by the mechanism are special objects that are created from the result of data analysis and allow you to automatically perform a forecast for new data in the future.
For example, an association search forecast model, built by analyzing customer purchases, can be used when working with a purchasing customer in order to offer him goods that he is likely to purchase along with the goods he has chosen.
The data analysis and forecasting mechanism provides users (economists, analysts, etc.) with the opportunity to search for non-obvious patterns in the data accumulated in the information base. This mechanism allows:
- search for patterns in the source data of the information base;
- manage the parameters of the analysis performed both programmatically and interactively;
- provide programmatic access to the analysis result;
- automatically display the analysis result in a spreadsheet document;
- create forecast models that allow you to automatically predict subsequent events or the values of certain characteristics of new objects.
The data analysis mechanism is a set of built-in language objects that interact with each other, which allows the developer to use its components in any combination in any application solution. Built-in objects make it easy to organize interactive configuration of analysis parameters by the user, and also allow you to display the analysis result in a form convenient for display in a spreadsheet document.
The mechanism allows you to work both with data received from the information base and with data received from an external source, pre-loaded into a table of values or a spreadsheet document:
By applying one of the types of analysis to the source data, you can obtain the result of the analysis. The result of the analysis represents a certain model of data behavior. The result of the analysis can be displayed in the final document, or saved for future use.
The further use of the analysis result is that on its basis a forecast model can be created that allows one to predict the behavior of new data in accordance with the existing model.
For example, you can analyze which products are purchased together (in one invoice) and save this analysis result in the database. Later, when creating the next invoice:
Based on the saved analysis result, you can build a forecast model, feed it “input” with new data contained in this invoice, and “output” receive a forecast - a list of goods that the counterparty B.S. Petrov. He will also most likely acquire them if they are offered to him:
The data analysis and forecasting engine implements several types of data analysis:
Implemented analysis types
general Statistics
It is a mechanism for collecting information about the data in the sample being studied. This type of analysis is intended for preliminary investigation of the data source being analyzed.
The analysis reveals a number of characteristics of numeric and continuous fields. When outputting a report to a spreadsheet document, pie charts are filled in to display the composition of the fields.
Search for associations
This type of analysis searches for frequently occurring groups of objects or characteristic values together, and also searches for association rules. Association search can be used, for example, to determine frequently purchased goods or services together:
This type of analysis can work with hierarchical data, which allows, for example, to find rules not only for specific products, but also for their groups. An important feature of this type of analysis is the ability to work both with an object data source, in which each column contains some characteristic of the object, and with an event source, where the characteristics of the object are located in one column.
To make the result easier to perceive, a mechanism for cutting off redundant rules is provided.
Sequence Search
The sequence search type of analysis allows you to identify sequential chains of events in a data source. For example, this could be a chain of goods or services that customers often purchase sequentially:
This type of analysis allows for hierarchical searches, which makes it possible to track not only the sequences of specific events, but also the sequences of parent groups.
A set of analysis parameters allows a specialist to limit the time distances between elements of the searched sequences, as well as adjust the accuracy of the results obtained.
Cluster analysis
Cluster analysis allows you to divide the original set of objects under study into groups of objects, so that each object is more similar to objects from its group than to objects from other groups. By further analyzing the resulting groups, called clusters, you can determine how this or that group is characterized and decide on methods for working with objects of various groups. For example, using cluster analysis, you can divide the clients with whom a company works into groups in order to apply different strategies when working with them:
Using the parameters of cluster analysis, the analyst can configure the algorithm by which the partitioning will be carried out, and can also dynamically change the composition of the characteristics taken into account in the analysis and configure weighting coefficients for them.
The result of clustering can be displayed in a dendrogram - a special object designed to display sequential relationships between objects.
Decision tree
The decision tree type of analysis allows you to build a hierarchical structure of classifying rules, presented in the form of a tree.
To build a decision tree, you need to select a target attribute on which the classifier will be built and a number of input attributes that will be used to create rules. The target attribute may contain, for example, information about whether the client switched to another service provider, whether the transaction was successful, whether the work was done well, etc. Input attributes, for example, can be the employee’s age, his work experience, the client’s financial condition, the number of employees in the company, etc.
The result of the analysis is presented in the form of a tree, each node of which contains a certain condition. To decide which class a new object should be assigned to, it is necessary, by answering questions at the nodes, to go through the chain from the root to the leaf of the tree, moving to the child nodes in the case of an affirmative answer and to the neighboring node in the case of a negative answer.
A set of analysis parameters allows you to adjust the accuracy of the resulting tree:
Forecast models
Forecast models created by the mechanism are special objects that are created from the result of data analysis and allow you to automatically perform a forecast for new data in the future.
For example, an association search forecast model, built by analyzing customer purchases, can be used when working with a purchasing customer in order to offer him goods that he is likely to purchase along with the goods he has chosen.
Using the data analysis mechanism in application solutions
To familiarize developers of application solutions with the data analysis mechanism, a demonstration information base is placed on the “Information and Technology Support” (ITS) disk. It includes a universal processing “Data Analysis Console”, which allows you to perform data analysis in any application solution, without modifying the configuration.
Modern information technologies/3. Software
Ph.D. Zhunusov K. M.
Kostanay State University named after A. Baitursynov
Formation of mechanisms for data analysis and forecasting
on the 1C: Enterprise platform
At the core of the forecasting process is the development of an economic forecast. It represents a scientifically based judgment about the possible states of an object in the future or about alternative ways and timing of achieving these states. In other words, this is an attempt to look into the future, predict it, foresee the state of the object under study after a certain period of time.
Forecasting is closely related to planning. There is a fairly common aphorism in scientific circles: “A forecast without a plan is a literary endeavor, a plan without a forecast is an administrative action.”
Common methods for both forecasting and planning are calculation-analytical, economic-statistical methods and economic-mathematical modeling.
Mechanisms for data analysis and forecasting as part of 1C Enterprise provide users (economists, analysts, etc.) with the ability to search for non-obvious patterns in the data accumulated in the information base and allow them to perform the following operations:
Search for patterns in the source data of the information base;
Manage the parameters of the analysis performed both programmatically and interactively;
Software access to the analysis result;
Automatic output of analysis results to a spreadsheet document;
Creation of forecast models that allow you to automatically predict subsequent events or the values of certain characteristics of new objects.
Data analysis and forecasting mechanisms are a set of built-in language objects that interact with each other, which allows the developer to use its components in any combination in any application solution. Built-in objects make it easy to organize interactive configuration of analysis parameters by the user, as well as display the analysis result in a form convenient for display in a spreadsheet document, in accordance with Figure 1. It is also important that the mechanism can work with data obtained both from the 1C information base and and from external sources (in the latter case, pre-loaded into a table of values or a spreadsheet document).
Picture 1. General scheme functioning of the data mining and forecasting mechanism
By applying one of the types of analysis to the source data, you can obtain a result that represents a certain model of data behavior. The result of the analysis can be displayed in the final document or saved for later use (based on it, a forecast model can be created that allows you to predict the behavior of new data).
One of the main trends in the accounting and management systems market is the constant increase in demand for the use of analytical data processing tools that ensure informed management decisions. However, today's customers are no longer satisfied with using traditional tools that allow them to create a variety of reports, pivot tables and charts based on predefined metrics and relationships analyzed manually. Enterprises increasingly need qualitatively different tools that allow them to automatically search for non-obvious rules and identify unknown patterns, which makes it possible to obtainnew knowledge based on the information accumulated by the company and sometimes make completely non-trivial decisions to improve business efficiency based on data mining methods.
Literature:
1 Glushchenko V.V. Forecasting. - M.: University Book, 2005.
2 Dubrova T. A. Methodological issues of production forecasting the most important species industrial products // Questions of statistics. -2004. -No. 1.-S. 52-57.
3 Radchenko M.G., Khrustaleva E.Yu. Tools for creating replicated applications "1C:Enterprise 8.2". - M.: Publishing house "1C-Publishing", 2011.
The mechanism is represented by a set of objects of the built-in 1C:Enterprise language. The interaction diagram of the main objects of the mechanism is shown in the figure. Setting up data analysis columns – a set of settings for input data analysis columns. For each column, the type of data contained in it is indicated, the role performed by the column, additional settings , depending on the type of analysis performed. Data analysis parameters – a set of parameters for the data analysis performed. The composition of the parameters depends on the type of analysis. For example, for cluster analysis, the number of clusters into which the original objects must be divided, the type of measurement of the distance between objects, etc. are indicated. Initial data is the source of data for analysis. The data source can be the result of a query, a cell area of a spreadsheet document, or a table of values. Analyzer is an object that directly performs data analysis. The object is given a data source and parameters are specified. The result of the work of this object is the result of data analysis, the type of which depends on the type of analysis. The result of data analysis is a special object containing information about the result of the analysis. Each type of analysis has its own result. For example, the result of data analysis - a decision tree - will be an object of the type DataAnalysisResultDecisionTree. In the future, the result can be displayed in a spreadsheet document using the data analysis report builder, can be displayed through programmatic access to its content, and can be used to create a forecast model. Any data analysis result can be saved for later use. A forecast model is a special object that allows you to make a forecast based on input data. The type of model depends on the type of data analysis. For example, a model created for data analysis - searching for associations will have the type Association Search Forecast Model. The source of data for the forecast is passed to the input of the forecast model. The result is a table of values containing predicted values. A sample for a forecast is a table of values, a query result, or an area of a spreadsheet document containing information on which to build a forecast. For example, for a forecast model - searching for associations, the selection may contain a list of products in a sales document. The result of the model’s work can recommend what products can still be offered to the buyer. Setting up sample columns is a set of special objects that show the correspondence between the forecast model columns and the forecast sample columns. Setting up result columns - allows you to control which columns will be placed in the result table of the forecast model. The result of the model is a table of values, consisting of columns, as specified in the settings of the resulting columns and containing the predicted data. The specific content is determined by the type of analysis. Data analysis report builder is an object that allows you to display a report on the result of data analysis. In addition, the report builder provides special objects for connecting with data in order to allow the user to interactively manage analysis parameters, setting up data source columns, setting up forecast model columns, etc. Types of analysis The mechanism allows you to perform the following types of analysis:- general Statistics
- Search for associations
- Sequence Search
- Decision tree
- Cluster analysis
- Data analysis – an object that performs data analysis. For it, you need to specify the data source and the necessary parameters for analysis.
- The result of data analysis is an object that is the result of data analysis work.
- Forecast model – created based on the result of data analysis. The object is the final link in the 1C analysis mechanism and generates a table of values that contains predicted values.
- General Statistics – This type of analysis is a simple statistical sampling of a data source. An example of application is analysis of sales by item for a period. The result of the analysis will be information about how much of a particular product was sold. The system will also calculate specific fields - maximum, minimum, median, average, range, standard deviation, number of values, number of unique values, mode.
- Search for associations – this type of analysis is designed to search for combinations that often occur together. Very good for finding items that are often purchased together. As a result of the analysis, the system will generate the following information: information about the processed data, associative groups, associative rules by which the groups are compared.
- Search for sequences - analysis that allows you to identify patterns in the analyzed data and offer further predictions. As a result of the analysis, the system will display information about the possibility of occurrence of certain events in percentage terms.
One of the main trends in the accounting and management systems market is the constant increase in demand for the use of analytical data processing tools that ensure informed decision-making. That is why one of the strategic directions for the development of the 1C:Enterprise software system has become the constant expansion of economic and analytical reporting capabilities. However, today's customers are no longer satisfied with traditional tools that allow them to generate a variety of reports, pivot tables and charts that are created based on predefined indicators and relationships and which must be analyzed manually. Enterprises increasingly need qualitatively different tools that allow them to automatically search for non-obvious rules and identify unknown patterns (Fig. 1). This is how you can generate qualitatively new knowledge based on the information accumulated by the company and sometimes make completely non-trivial decisions to improve business efficiency, using data mining (DAM) methods.
Rice. 1. The logic of developing the “intelligence” of solved analytical problems. Released in summer 2003 new version technology platform "1C:Enterprise 8.0" made it possible to significantly expand the capabilities of business analytics in the system (see sidebar). However, there is one thing you need to do here important note. The 1C platform software develops not only in “steps”, from version to version, but is constantly improved and expanded within one version, and in two directions - technological and applied. So, after the first announcement of the G8, more than a dozen platform releases have already been released, latest version (as of January 2006) is numbered 8.0.13, and it is quite different from what it was two and a half years ago! One of the areas of development of "1C:Enterprise 8.0" is precisely the mechanisms of business analytics; in particular, IAD tools appeared in it only in 2005. It is important to note that most analysis functions are implemented at the technology platform level and become available to users only after inclusion in new releases of application solutions. Thus, there is some gap (sometimes several months) between the emergence of new features and their provision to users. With this problem in mind, in order to bridge the gap, 1C released in September 2005 a special application solution “Data Analysis Subsystem” (DAS), which can be built into any configuration of the 1C:Enterprise 8.0 platform. In addition to a wide range of basic functions, the package includes more than 30 pre-configured models for a typical Trade Management configuration. PAD includes those qualitatively new IAD tools that were previously absent from 1C programs. To directly analyze and predict data, specific skills and knowledge are not required. A good command of the analyzed subject area and an understanding of the main cause-and-effect relationships in it are assumed. Preparing data sources and predictive models requires the ability to use the query builder and knowledge of how to place information in configuration metadata objects. IAD algorithms, included in the new configuration (version 1.0.5), form analytical models (templates) that describe patterns in the source data. These models are of independent value (they can be used repeatedly), and are also used for automated generation of forecasts, including scenario ones, with previously unknown indicators (Fig. 2). The IAD mechanism is a set of built-in language objects that interact with each other, thanks to which the developer can use its components in any combination in any application solution. Built-in objects make it easy to organize interactive configuration of analysis parameters by the user, as well as display the analysis result in an easy-to-display form in a spreadsheet document. By applying one type of analysis to the source data, you can obtain a result that will represent a certain model of data behavior. The result of the analysis can be displayed in the final document or saved for later use - based on it, you can create a forecast model that allows you to predict the behavior of new data.
Rice. 2. General scheme of functioning of the data mining mechanism. The current version of the subsystem implements the methods that have received the greatest commercial distribution in world practice, namely:
- clustering - implements grouping of objects, maximizing intragroup similarity and intergroup differences;
- decision tree - provides the construction of a cause-and-effect hierarchy of conditions leading to certain decisions;
- search for associations - searches for stable combinations of elements in events or objects.
Clustering
The purpose of clustering is to select from a set of objects of the same nature a certain number of relatively homogeneous groups (segments or clusters). Objects are distributed into groups in such a way that intragroup differences are minimal, and intergroup differences are maximum (Fig. 3). Clustering methods make it possible to move from an object-by-object to a group representation of a collection of arbitrary objects, which significantly simplifies their handling. Below are a few possible scenarios application of clustering in practice. Customer segmentation based on a certain set of parameters, it makes it possible to identify stable groups among them that have similar purchasing preferences, sales levels and solvency, which greatly simplifies the management of customer relationships. At classification of goods Quite conventional classification principles are very often used. Isolating segments based on a group of formal criteria makes it possible to identify truly homogeneous groups of goods. In the context of a wide and rather heterogeneous product range, assortment management at the segment level, compared to product-level management, significantly increases the efficiency of promotion, pricing, merchandising, and supply chain management. Manager segmentation allows you to more effectively plan organizational changes, improve motivational schemes, adjust the requirements for hired personnel, which ultimately allows you to increase the manageability of the company and the stability of the business as a whole.Rice. 3. Data analysis using clustering method. The similarity and difference between objects is determined by the “distance” between them in the space of factors. The method for measuring distance depends on the metric, which indicates the principle for determining the similarity/difference between sample objects. The current implementation supports the following metrics:
- "Euclidean metric" is the standard distance between two points in N-dimensional Euclidean attribute space;
- “Euclidean metric squared” - enhances the influence of the difference (distance) on the clustering result;
- "city metric" - reduces the impact of emissions;
- “dominance metric” - defines the difference between sample objects as the maximum existing difference between the values of their attributes, therefore it is useful for enhancing the differences between objects for one attribute.
- “short-range communication” - the object joins the group for which the distance to the nearest object is minimal;
- “long-distance communication” - the object joins the group for which the distance to the farthest object is minimal;
- “center of gravity” - the object joins the group for which the distance to the center of the cluster is minimal;
- "k-means" method - arbitrary objects are selected, which are considered cluster centers, then all analyzed objects are sequentially sorted out and joined to the cluster closest to them. After attaching the object, it is calculated new center cluster, which is calculated as the average value of the attributes of all objects included in the cluster. The procedure is repeated as long as the cluster centers change.
- cluster centers, which are a set of averaged values of the input columns in each cluster;
- a table of intercluster distances (distances between cluster centers), which determine the degree of difference between them;
- values of forecast columns for each cluster;
- rating of factors and a tree of conditions that determined the distribution of objects into clusters.
Search for associations
This method is designed to identify stable combinations of elements in certain events or objects. The results of the analysis are presented in the form of groups of associated elements. Here, in addition to the identified stable combinations of elements, detailed analytics on associated elements are provided (Fig. 4).Rice. 4. Presentation of the results of the analysis by the “search for associations” method in the form of groups of associated elements. The method was originally developed to find typical combinations of items in purchases, which is why it is sometimes also called shopping basket analysis. In this scenario, the associated elements are usually product groups or individual products. And the grouping object that combines the elements of the samples can be any object of the information system that identifies the transaction: for example, a buyer’s order, an act for the provision of services, or a cash receipt. Information about patterns in customer product preferences increases the efficiency of customer relationship management (in terms of advertising campaigns and marketing promotions), pricing (formation of complex offers and discount systems), inventory management and merchandising (distribution of goods in sales areas). Another example of using this method is to determine customers' preferred combinations of advertising channels to avoid duplication when running targeted advertising campaigns. This allows you to significantly reduce the costs of such events. The association search algorithm implemented in the platform has quite flexible means of controlling the adequacy of analysis or forecast models. The "Minimum percentage of cases" parameter determines the "triggering threshold" of the algorithm for a particular combination of elements in an event or object, which allows you to ignore weakly common associations. The "Minimum Reliability" parameter determines the required stability of the sought associations, and the "Minimum Significance" parameter allows you to identify the highest priority ones. The “Rules cutoff type” parameter greatly facilitates the perception of analysis and forecast results, which can take the values “Cut off redundant” and “Cut off those covered by other rules.” For the practical interpretation of the results obtained using this algorithm, it is critically important to split the initial set of associated elements into groups that are truly homogeneous from the point of view of the analysis being carried out.
Decision tree
As a result of applying this method to the source data, a hierarchical (tree-like) structure of rules of the form “if... then...” is created, and the analysis algorithm ensures the identification of the most significant conditions and transitions between them at each stage. This algorithm is most widely used in identifying cause-and-effect relationships in data and describing behavioral patterns. A typical area of application of decision trees is the assessment of various risks, for example, the closing of an order by a client or its transfer to a competitor, untimely delivery of goods by a supplier or late payment of a trade loan (Fig. 5). Typical input factors of the model are the amount and composition of the order, the current balance of mutual settlements, credit limit, prepayment percentage, delivery conditions and other parameters characterizing the forecast object. Adequate risk assessment ensures informed decisions are made to optimize the return/risk ratio of a company's activities, and is also useful for increasing the realism of various budgets.Rice. 5. The use of the “decision tree” method allows, based on the input factors of the model (a), to obtain an assessment of the risks of accepting certain management decisions(b). An example illustrating the algorithm’s ability to identify cause-and-effect relationships is the task of optimizing the work of the sales department. To solve it, we will choose an indicator of the effectiveness of sales managers, for example, specific profitability per client, as a predicted value, and as factors - a set of data that potentially influences the result. The algorithm will determine the factors influencing greatest influence on the result, as well as typical combinations of conditions leading to a particular result. Moreover, the “Data Analysis” subsystem will allow you to estimate (predict) the expected values of the target indicator based on current data, as well as make a “what if ...” forecast by changing the indicators supplied to the model input. The results of analysis and forecast using decision trees can significantly reduce the impact of the uncertainty of the business environment on the state of the company, as well as solve a wide range of problems related to identifying complex and non-obvious cause-and-effect relationships. The Decision Tree algorithm forms a cause-and-effect hierarchy of conditions leading to certain decisions. As a result of applying this method to the training sample, a hierarchical (tree-like) structure of splitting rules of the “if... then...” type is created. The analysis algorithm (model training) comes down to an iterative process of identifying the most significant conditions and transitions between them. Conditions can be both quantitative and qualitative in nature and form the “branches” of this abstract tree. Its “foliage” is formed by the values of the predicted attribute (decision), which, like the transition conditions, allow both qualitative and quantitative interpretation. The combination of these conditions imposed on the factors and the structure of transitions between them to the final solution form the forecast model. This algorithm has become most widespread in assessing the outcomes of various event chains and identifying cause-and-effect relationships in samples. The significance and reliability of the model of this algorithm is controlled using the parameters “Simplification type”, “Maximum tree depth” and “Minimum number of elements in a node”. The results of sample analysis using the “Decision Tree” algorithm are:
- rating of factors, which is a list of factors that influenced the decision, sorted in descending order of importance (“citations” in tree nodes);
- comparison of decisions (values of the forecast column) and the conditions that determined them, in other words, the “Effect-Cause” tree;
- “Cause-Effect” tree, which is a set of transitions between conditions that determines a particular decision (essentially, a visual representation of the forecast model).
| Joint solutions "1C" In addition to the functions implemented directly within the framework of the 1C:Enterprise 8.0 platform, the arsenal of 1C business analytics tools is replenished with specialized solutions created, among other things, within the framework of the 1C-Joint project (http://v8.1c.ru/ solutions) - with the participation of the company's partners and independent developers (see "Joint solutions of 1C and its partners", "BYTE / Russia" No. 9 "2005). Here we note two products related to the use of intelligent analysis methods - This is "1C:Enterprise 8.0. 1C-VIP Anatech: ABIS. ABC. Management accounting and cost calculation" (developer partner - consulting company "VIP Anatech") and "1C-VIP Anatech-VDGB: ABIS. B.S.C. Balanced Scorecard" (partner-developers - VIP Anatech and VDGB companies). |
Typical business scenarios for using IAD methods
The PAD documentation has a section devoted to typical examples of the use of data mining in relation to the "1C: Trade Management 8.0." configuration. Here we present several such business scenarios.Customer Relationship Management
Scenario "Planning" advertising campaign" Planning the upcoming advertising campaign is considered from the point of view of optimizing the distribution of the allocated budget across advertising channels, based on regional, product, customer and other indicators of the target segment, as well as on the effectiveness of advertising channels in the specified sections in some previous planned period. Algorithm- "Cluster analysis". Forecast Attributes- the share of responses to the advertising channel of conditionally homogeneous segments identified by the algorithm. Calculated Columns- the share of advertising channels in the advertising campaign budget, taking into account the likely share of responses and effectiveness (in terms of resulting revenue) of each advertising channel. Example of a pattern: Class A clients of region P, who prefer product group P, are attracted by the same advertising channel as clients of region N, who prefer product group Y.Supply chain management
Scenario "Optimization of supplier selection by product group" The selection of dominant first-line suppliers for key product groups is extremely important for stabilizing the logistics system in particular and the overall supply chain management system in general, as well as reducing the average duration of supply chains. On the other hand, closer integration with major suppliers usually allows for significant reductions in the cost of goods. In this regard, it is of interest to analyze stable combinations of suppliers in various product groups in comparison with analytics for suppliers associated within the groups. This allows you to identify “intersections” of suppliers in various product groups and optimize relationships with them. Algorithm- "Search for associations." Forecast Attributes- sustainable combinations of suppliers. Main Factors- product groups. Decoding- analytics on suppliers (volume of purchases, revenue, terms of delivery and payment, order completion times - pessimistic, optimistic, average). Example of a pattern: a stable association of a large and unpredictable supplier A and a predictable medium-sized supplier B in a large number of product groups. When placing orders for competitive product groups, it is possible to position a medium-sized supplier as the main one, if the volume of the order for a large one does not exceed a certain threshold (giving a significant gain in scale).Personnel Management
Scenario: Profiling Sales Managers by Key Performance Indicators Determining the effectiveness of managers (retention, customer search, communication efficiency, collection of conditional and unconditional receivables, specific performance indicators per client, etc.) is of interest not only from the point of view of creating a system of material incentives for managers, but also from the point of view of effective standardization parameters of their activities. Algorithm- "Decision trees". Forecast Attributes- key performance indicators of the sales department (number of key customers, churn and acquisition rates, lost income per month, attracted income per month, income per month per client, total revenue from clients, etc.). Main Factors- number of active clients, revenue, income, specific indicators per client, communication efficiency. Depending on the predictive attributes, the composition of factors can vary significantly. Example of a pattern: managers who provide the best indicators of collection of receivables (ratio of DS receipts to revenue) have a retention ratio > 0.8; attraction coefficient > 0.25; the number of simultaneously open transactions is no more than 15, but not less than 10; the intensity of events per day is no more than 10, but not less than 3; the number of active clients in the period is at least 50, but not more than 100.Conclusion
Modern business is so multifaceted that the factors potentially influencing a particular decision can number in the dozens. Competition is increasing day by day, life cycle products are becoming shorter, customer preferences are changing faster and faster. To develop a business, it is necessary to respond as dynamically as possible to the rapidly changing business environment, taking into account the subtle and sometimes elusive patterns of developments. Which customer groups will respond to the promotion, and which will irrevocably go to competitors? Should I open a new business line or hold off for now? Will the buyer be late in payment or the supplier late in shipment? What are the opportunities for growth and where are the potential threats? Thousands of managers ask themselves and their colleagues these questions every day. The data analysis subsystem implemented in the 1C:Enterprise 8.0 platform is designed to help users of the corporate information system quickly find answers to non-trivial questions, providing automated transformation of data accumulated in information system, into practical and well-interpretable patterns.Economic and analytical reporting in "1C:Enterprise 8.0"The 1C:Enterprise 8.0 platform includes a number of mechanisms for generating economic and analytical reporting that allow you to generate interactive documents (and not just printed forms) within the framework of certain application solutions. Thus, the user can work with reports in the same way as with any screen form, including changing report parameters, rebuilding it, using “decoding” (obtaining additional reports based on individual elements of an already generated report), etc. In addition , there are several universal software tools that allow you to generate any arbitrary reports, depending on the tasks. This can also be done by the users themselves (sufficiently experienced) who are well familiar with the structure of the application solution being used. Below we will briefly look at the main reporting tools in 1C:Enterprise 8.0. Requests- this is one of the ways to access data in “1C:Enterprise 8.0”, with the help of which information is retrieved from the database according to certain conditions, usually in combination with the simplest processing of the received data: group, sort, calculate. Changing data using queries is impossible, since they were originally designed to quickly obtain information from large amounts of information. The database is implemented as a set of interconnected tables that can be accessed either individually or several tables in conjunction. To implement his own algorithms, the developer can use a query language based on SQL and containing many extensions that reflect the specifics of financial and economic problems and reduce the effort spent on creating application solutions. The platform includes a query designer that allows you to compose the correct query text using only visual tools (Fig. 6).Rice. 6. The query builder (a) allows the developer to compose the query text (b) exclusively by visual means. spreadsheet document is a powerful mechanism for visualizing and editing information, including using dynamic reading of information from a database. A spreadsheet document can be used on its own or be part of any of the forms used in the application solution. At its core it resembles spreadsheets(consists of rows and columns in which data is placed), but its capabilities are much wider. It supports grouping, decryption, and annotation operations. In a document you can use different kinds report design, including graphic diagrams. A tabular document can contain pivot tables, which themselves serve as an effective tool for programmatically and interactively presenting multidimensional data. Output form constructor helps the developer create reports and present report data in a convenient tabular or graphical form. It includes all the features of the query designer, as well as form creation and customization. Report Builder is an object of the built-in language that provides the ability to dynamically create a report both programmatically and interactively (Fig. 7). Its operation is based on a request, by which the user is given the opportunity to interactively configure all the main parameters contained in the request text. The results of this query are output to a spreadsheet document, which can also use information from arbitrary data sources. The developer, using the report builder commands, can change the parameters available to the user for configuration. |
