ANALLYZ

Smoothing

Decomposition Options

ARIMA Options

Continuous Data Options
Process Capability Options
Attribute Data Options
Input minimum Support and Confidence for Association Rules
Analyze Rules for an item
Direction for Association rule for an item
Input minimum Support and Confidence for Association Rules
Select Initial State from the drop-down list
Input number of Time steps ahead for analysis
Press VIEW to make selections or view results
Download the results
Save Forecast with Decomposition
Save Forecast with ARIMA
Data View

The Raw Data is presented in the form of a Table, wherein the data can be sorted by any Column, or any value can be searched for. In addition to that, a Histogram is also generated for the variable of intereset. We should use a Normally distributed data for Forecasting and Process Excellence . It may be useful to assess presence of outliers. In case an error message appears, CHANGE option can be used to regenerate the chart.

Forecasting

The prime application of Forecasting is in Supply Chain Managemenet for Demand Forecasting. In addition to that, it is extensively used for forecasting Commodity Prices, in Financial Market Analysis and in Economics. Time Series Analysis plays a crucial role in these areas, not only in analyzing the Stable Components - Trend, Cyclic and Seasonal, but also using them for forecasting.

Simple and Exponential Moving Averages are Smoothing Techniques which help find Stable components like Trend, Cyclic and Seasonal etc. The Order of Smoothing should be chosen for Least Error displayed by the Bar Chart. Alpha used in Exponential MA, is automatically calculated for least error, and no explicit input is needed. These techniques are primarily exploratory in nature, and not meant for Forecasting.

Decomposition is nothing but Triple Exponential Moving Average, which is also known as Holt-Winters Method. Alpha, Beta and Gamma are optimized for least error, and no explicit value is needed. Frequency is mandatory input in this case. Frequency of Weekly Data is 52, of Monthly Data is 12 and of Quarterly Data is 4. Future Time is the number of Steps used to generate forecast. As an example, 20 would generate forecast for 20 weeks ahead. This is needed to generate Forecast.

ARIMA Forecast needs AR, I and MA parameters. One can start with an arbitrary value of 1, 0, 1 respectively and subsequently improve them based on the ARIMA Output generated.


                

Statistical Process Control

XBar Chart plots the process mean over time for Continuous data in subgroups. This chart is widely used to examine the stability of processes in many areas. In the downloadable Example Data, it is length of a tube which is being analyzed. At any point in time, the data is collected in samples, which are known as subgroups. For example, one can use XBar Charts to monitor the process mean for subgroups of Diameter of a Part, Time taken to resolve a Customer Call, or Demand for a Product over time.

XBar.One Chart is very similar to XBar Chart, however in this case the data charted are simply the sample values, and the within-group standard deviation is estimated by moving ranges of k(by default 2) points.

R Chart plots the process range over time for variables data in subgroups. This control chart is widely used to examine the stability of processes in many industries.

XBar and R(or S)Chart should be inspected simultaneously. XBar plots the Central Values of Sample Measurements and R Chart plots the variations(Ranges) for these samples. We should inspect the variations with R chart before inspecting average with an Xbar chart.

S Chart is similar to R Chart and plots Std. Deviations in place of Ranges. R chart is used when subgroup size is 8 or less. S chart is used when subgroup size is 9 or more.

Process Capability compares the Process Spread to the Specification Spread. Hence it needs two inputs UCL and LCL, which are Upper Specification and Lower Specification Limits respectively. The comparison is made by forming the ratio of the spread between the Process Specifications (UCL and LCL) to the spread of the process values, as measured by 6 process standard deviation units (the Process Spread).

An Upper Control Limit (UCL) and Lower Control Limit (LCL) is mandatory for Process Capability Analysis

Process Capability Analysis is only valid for In-Control Processes which are inspected through XBar, R or S Charts. The Processes which do not show any non-random trends are considered In-control Processes. Two types of non-random values are considered. The Mean or Range may cross the Upper and Lower Control Limits. Also the Values may exhibit constant increasing(or decreasing) trends, showing clusters or hugging the limits.



p Chart is used for Attribute Data, where the measurements is in terms of classes like defective/non-defective, pass/fail etc. and not on an absolute number. P Chart is to monitor the proportion of non-conforming observations in a sample. Hence in a sample size of 50, if 5 are defectives then p value would be 0.1. np Chart is used instead for number of non-confirming observations. c and u charts are used for defectives per unit and use Poisson Distributions.

In case of Continuous Data, the Measurement and Subgroup(Sample) identifier columns need to be selected. Please download Continuous_data.csv file to run the example. In case of Attribute Data, the column identifying Class of Interest and the Sample Size needs to be selected. Please download Attribute_data.csv file to run the example.

Out-of-Control Process would reveal observations beyond limit(red color) and violating runs(yellow color) showing non-random trends. In Six Sigma Terminology these are called Assignable Cause variations or Special Cause variations. This needs a Root Cause Analysis and the process needs to be brought in control.

Association Rules

Association Rule Mining can provide three results

Frequent Item Sets

The most frequent items or item sets are displayed in the form of a Bar Plot. The default number of item sets to be displayed is 5, which can be increased.


Overall Strong Rules

The strong rules can be viewed by providing a Minimum Support and Confidence. The rules are sorted in decreasing order in terms of Confidence. By default top 5 rules are displayed, which can be increased(or decreased). For large data sets, a very low Support may be required. However in few cases,especially for small data sets a very low Support may give an error message and it may warrant increasing Support to a higer value.

Although minimum Confidence is set at 0.8, it may need to be decreased in certain cases, if an error message is encountered.



              

Specific Rules for a Frequent Item

This analysis displays the results for a certain item. If association rules for a particular item of interest is desired, then that Item should be typed in the Text Box. As an example, whole milk(case sensitive) can be entered to see all the association rules for this item. The rules can be viewed in two ways. Before refers to rule where item sets before the input Item occur. As an example, if item1 & item2 then whole milk refers to Before or LHS rules. On the other hand, after rules refer to item sets occuring after the input Item. As an example, if whole milk then item1, item2 is a After or RHS rule.As earlier number of rules, minimum support and confidence should be changed and adjusted accordingly in case of error messages.

Association Rule Analysis needs data in a specific format. Please observe the format of the Example file.



              

Markov Chain Analysis


The Column Chart below shows the probabilities for all the States at all future time steps. As an example, if one wants to see the probabilities of a Customer shopping at Retailers A, B and C five months from current State (0,B,0), then probabilities for A, B and C for months 1 to 5 are displayed. The state with the highest probability can be considered the most likely event at a particular time step. In a similar fashion the migration of a Bond from one rating to other ratings for multiple years can be visualized.