Management Report: Association Rule Mining and Data Visualization

This report explores the application of association rule mining in a business context, specifically focusing on an online food delivery company. It explains the benefits of association rule mining, such as identifying patterns in customer transactions to offer targeted promotions and improve revenue. The report delves into the mathematical concepts behind association rules, including support, confidence, lift, and conviction, and highlights the importance of considering these metrics for accurate insights. The report also addresses the generation of association rules, including frequent itemset identification using algorithms like Apriori. Furthermore, the report includes an analysis of executive/performance dashboards and recommends metrics for monitoring organizational performance, along with a paper prototype. Finally, the report suggests the creation of data visualizations using tools like PowerBI, based on the dataset utilized in the previous assignments.

Running head: TOPICS IN MANAGEMENT
TOPICS IN MANAGEMENT
Name of the Student
Name of the University
Author Note

1TOPICS IN MANAGEMENT
1.
Association rule mining is widely used in data mining applications by various type of
organization. This process is widely preferred for discovering patterns between the variables
which apparently look unrelated. Most of the machine learning algorithms are good for
dealing with numeric data, however, the association rule mining is particularly good for non-
numeric data like nominal variable or strings (Wang and Zheng 2020). The algorithm of this
technique is just a bit more than the simple counting process. Let’s consider an online food
delivery company that want to find patterns in customers’ transactions of their products.
Now, the customer transactions can be classified by the association rule where frequent
occurring patterns, association and correlation between the different item transactions can be
found by simple if-then statements. The ‘if’ of association is expressed as independent
variable which is also known as antecedent of the association and ‘then’ is expressed as
dependent variable which is the ‘consequent’ of the association. For example if it is found
from the collected data of transaction of the company that 70% of the customers who ordered
bread from the company also ordered milk then by the association rule ‘if bread is ordered
then it is 70% likely to order milk’ (Haraty and Nasrallah 2019). In this statement the bread is
the antecedent and milk is the consequent. This can used by the online food company to
promote milk when bread is ordered by a customer as a suggestion. This algorithm not only
helps the food company to better understand the interests of customer but also helps to earn
more revenue and thus profit for the company. The association rules are basically based on
the any type of raw data and association is performed by looking at frequent if/then patterns
between the variables. The importance of the relationship in association is understood by
observing the two parameters, support and confidence respectively. The parameter support
tells the frequency of if/then patterns between the variables and confidence tells the
frequency of True if-then instances. Hence, by association rule the attempt is to find the

2TOPICS IN MANAGEMENT
relationship between the variable by measuring the frequency and thus it is attempted to find
the reason behind buying two or more items at once by the customers of the food company.
In association rule a mathematical term lift is calculated which indicates the power of
association. The lift is defined by the ratio of observed frequency of occurrence of true
relations and the expected frequency of true relations. Lift if simple calculated by the
counting the transactions in the database and then performing simple operations of
mathematics. Hence, mathematically the lift is the joint probability of two items y and x
which is divided by their corresponding probabilities product. Hence, lift = P(x,y) / (P(x) *
P(y)), where x and y are two items. Now, if the two items are independent statistically then
P(x,y) = P(x) * P(y) which gives the Lift value 1. Thus if the items or variables are
independent then this can be indicated by the lift value or less the relation between the
variable the lift will be more close to 1 or can be referred as less power of dependency
between the variable by the rule of association. Also, it must be noted here that the if there is
anti-correlation between variables then also the lift values can be less than 1 such as for the
mutually exclusive items. For the food company there can be found significant correlation of
lift between veg and non-veg items as a person who is vegetarian is surely not non-vegetarian
and thus if veg items are selected then the non-veg items would not be selected and this is
anti-correlation. Thus the food company must not blindly go to conclusion with the results of
association rule algorithm but must investigate further into result to understand the relations.
Another important mathematical term in association rule is the conviction which is defined
between two items x and y as,
conv ( x=¿ y )= 1 – support ( y )
1−confidence ( x=¿ y )
As, an example for the food company if found that if the milk and bread are bought together
then butter is also bought which is expressed by relation {milk, bread} => {butter} has

3TOPICS IN MANAGEMENT
conviction equal to the ratio of expected frequency of occurrence of X without Y when X and
Y which is divided by observed frequency occurrence of the same. If the value of conviction
is calculated to be 1.1 then it is interpreted as there is 10% more chance that the rule {milk,
bread} => {butter} is incorrect when the relation between X and Y are entirely random. Now,
for decisive evidence the association rule needs to satisfy the minimum support and
confidence values simultaneously (Datta and Bose 2016). The generation of association rule
is divided normally in two stages. In the first stage the frequent items or variables are found
by specifying a minimum support to the entire database of customer transaction. In the
second step rules are formed between these frequent variables that satisfies the minimum
confidence constraint. The second step is generally simple and all the critical things are
considered in the first step for most cases. This is because as the transaction dataset becomes
large the power set of all possible item set exponentially increases. When the dataset is of
size I then the power set of all possible item set that does not include the empty items has the
size if 2^n -1 where n is the number of items. Thus an efficient search algorithm must be
applied to find the support which is often known as downward-closure property or anti-
monotonicity that ensures that in a frequent item set all of its subsets must also be frequently
thus no unnecessary infrequent items are not included in the final item set. Many association
rules like Apriori and Eclat use this particular property to locate the frequent items from a
large dataset (Yuan 2017). Hence, the food company can also use either of the association
rule algorithm to segregate and find the useful patterns between the items ordered by their
customer in recent history of transaction. Thus it is fully recommended to use the association
rule algorithms and necessary software to find patterns of customers’ preferences and hence
to implement proper business decisions for better serving the people and to grow in their
business.

4TOPICS IN MANAGEMENT
References:
Datta, S. and Bose, S., 2016. Mining and ranking association rules in support, confidence,
correlation, and dissociation framework. In Proceedings of the 4th International Conference
on Frontiers in Intelligent Computing: Theory and Applications (FICTA) 2015 (pp. 141-152).
Springer, New Delhi.
Haraty, R.A. and Nasrallah, R., 2019. Indexing Arabic texts using association rule data
mining. Library Hi Tech.
Wang, C. and Zheng, X., 2020. Application of improved time series Apriori algorithm by
frequent itemsets in association rule data mining based on temporal constraint. Evolutionary
Intelligence, 13(1), pp.39-49.
Yuan, X., 2017, March. An improved Apriori algorithm for mining association rules. In AIP
conference proceedings (Vol. 1820, No. 1, p. 080005). AIP Publishing LLC.