FORM-2
THE PATENT ACT,1970
(39 OF 1970)
AND
THE PATENT RULES, 2003
(As Amended)
COMPLETE SPECIFICATION (See section 10;rule 13)
"PREDICTING COMPLAINING BEHAVIOUR OF A USER BASED ON MACHINE LEARNING TECHNIQUES"
ICICI Prudential Life Insurance Company Limited, a corporation organized and existing under the laws of India of ICICI Prulife Towers, 1089, Appasaheb Marathe Marg, Prabhadevi, Mumbai 400025, India
The following specification particularly describes the invention and the manner in which it is to be performed:

PREDICTING COMPLAINING BEHAVIOUR OF A USER BASED ON MACHINE LEARNING TECHNIQUES
TECHNICAL FIELD
The present invention relates to the field of machine learning, in particular to predicting complaining behavior of user based on machine learning techniques.
BACKGROUND
[0001] Machine learning techniques find their applications in various fields these days. One or more machine learning models can be trained using sample historical data so that the models can be used to learn based on the sample data. Based on the learning by the machine learning models, the models can be used to predict an outcome.
[0002] One such application of the machine learning technology can be found for predicting complaining behavior of users. In any service industry (for example, in the life insurance sector), user life cycle tends to be long and uniquely sensitive, therefore customer centricity becomes critical. One aspect of user centricity is a robust and responsive complaint redressal system – redressing complaints with minimum turn-around time and to the satisfaction of the user. Less complaints would mean enhanced user experience by the service providers.
[0003] A more proactive approach, however, is to identify pockets of users where complaints are more likely to arise and use this classification to pre-sensitize the service team about the user they are about to deal with. Therefore, the need arises to provide improved techniques to predict the user behavior and categorize the users based on their propensity to complain.

SUMMARY
[0004] The following presents a simplified summary of the subject matter in order to provide a basic understanding of some aspects of subject matter embodiments. This summary is not an extensive overview of the subject matter. It is not intended to identify key/critical elements of the embodiments or to delineate the scope of the subject matter.
[0005] Its sole purpose is to present some concepts of the subject matter in a simplified form as a prelude to the more detailed description that is presented later.
[0006] The primary objective of the present invention is to provide machine learning based techniques to categorize the users based on their probability to complain.
[0007] Another objective of the present invention is to provide the categorization of users to the customer touchpoints to ensure targeted interventions for customers having high propensity to complaint.
[0008] In one embodiment, a method for predicting the propensity to complain by a user is disclosed. The method comprises obtaining a first type of variables related to the user, obtaining a second type of variables user related to interaction with various user interaction platforms, obtaining a third type of variables indicating incidence of the user to complain, determining the multivariate relationship between first type of variables and third type of variable based on first logistic regression model to predict the first probability of complaining, determining the multivariate relationship between second type of variables and third type of variable based on second logistic regression model to predict the second probability of complaining, and classifying the user based on the first probability and the second probability of complaint.

[0009] In another embodiment, an apparatus for predicting the propensity to complain by a user is disclosed. The method comprises obtaining a first type of variables related to the user, obtaining a second type of variables user related to interaction with various user interaction platforms, obtaining a third type of variables indicating incidence of the user to complain, determining the multivariate relationship between first type of variables and third type of variable based on first logistic regression model to predict the first probability of complaining, determining the multivariate relationship between second type of variables and third type of variable based on second logistic regression model to predict the second probability of complaining, and classifying the user based on the first probability and the second probability of complaint.
[0010] These and other objects, embodiments and advantages of the present invention will become readily apparent to those skilled in the art from the following detailed description of the embodiments having reference to the attached figures, the invention not being limited to any particular embodiments disclosed.
BRIEF DESCRIPTION OF FIGURES
[0011] The foregoing and further objects, features and advantages of the present subject matter will become apparent from the following description of exemplary embodiments with reference to the accompanying drawings, wherein like numerals are used to represent like elements.
[0012] It is to be noted, however, that the appended drawings along with the reference numerals illustrate only typical embodiments of the present subject matter, and are therefore, not to be considered for limiting of its scope, for the subject matter may admit to other equally effective embodiments.

[0013] FIGURE 1 illustrates a simple block diagram of an apparatus for predicting the propensity to complain by a user, according to an embodiment of the present invention.
[0014] FIGURE 2 illustrates a simplified flowchart of a method for categorization of users based on their propensity to complain considering contract type as insurance policy of a user to complain, according to an embodiment of the present invention.
[0015] FIGURE 3 illustrates a flow chart of a method for predicting the propensity to complain by a user, according to an embodiment of the present invention.
[0016] FIGURE 4 is a block diagram illustrating an exemplary computing device in which one or more embodiments of the present invention may operate, according to an embodiment.
DETAILED DESCRIPTION
[0017] Exemplary embodiments now will be described with reference to the accompanying drawings. The disclosure may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey its scope to those skilled in the art. The terminology used in the detailed description of the particular exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting. In the drawings, like numbers refer to like elements.
[0018] It is to be noted, however, that the reference numerals used herein illustrate only typical embodiments of the present subject matter, and are therefore, not to be considered for limiting of its scope, for the subject matter may admit to other equally effective embodiments.

[0019] The specification may refer to “an”, “one” or “some” embodiment(s) in several locations. This does not necessarily imply that each such reference is to the same embodiment(s), or that the feature only applies to a single embodiment. Single features of different embodiments may also be combined to provide other embodiments.
[0020] As used herein, the singular forms “a”, “an” and “the” are intended to include the plural forms as well, unless expressly stated otherwise. It will be further understood that the terms “includes”, “comprises”, “including” and/or “comprising” when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
[0021] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
[0022] The figures depict a simplified structure only showing some elements and functional entities, all being logical units whose implementation may differ from what is shown. The connections shown are logical connections; the actual physical connections may be different. It is apparent to a person skilled in the art that the structure may also comprise other functions and structures.

[0023] The present invention proposes to predict the propensity of a user to
complain and also provide categorization of users based on their probability to
complain. This categorization of users is provided to a service team, thereby
ensuring that the team is pre-sensitized about the type of user that they are about to
5 interact with and ensure enhanced user experience. The various parameters which
can be used as input in various logistic regression models include the first type of variables and the second type of variables. The first type of variables include demographic and contract related indicators, and the second type of variables include previous interaction experience indicators of the user.
10
[0024] Models like logistic regression (LR) may be used to fit the data to predict if the user is going to complain. The model classifies users into different tagging (Red, Maroon, Amber, Yellow, Green – i.e., RMAYG classification) based on their probability to complain. Users with Red tagging have the highest propensity to
15 complain and those with Green least. After identifying the categories of users, the
RMAYG classification is provided to the customer touch points (for e.g., call center, branch, email, etc.). This ensures that the team is pre-sensitized about the type of user that they are about to interact with and ensure a better client experience.
20 [0025] Referring to Figure 1 now, a block diagram of an apparatus 100 for
predicting propensity to complain by a user is disclosed. The apparatus helps in determining complaining behavior of a user in advance and also to categorize the users based on their probability to complain. The system 100 comprises a receiving module 102, a processing module 104, a memory 106, and a user categorization
25 module 108. In one embodiment, the functioning of each of the modules may be
performed by a processor 408.
[0026] The proposed invention can be implemented in any service industry where
user interactions are involved and on different types of contracts. In the case of
30 service industry being insurance industry, the present invention can be implemented
on multiple types of policies, e.g., but not limited to the protection type insurance
7

policy and the saving type insurance policy. These insurance policies take into account different features of these two product classes (i.e., protection and savings).
[0027] Each of the different types of contracts (or policies in the case of insurance
5 industry) comprises of different types of variables. The receiving module 102
obtains a first type of variables and a second type of variables. In the case of
insurance policies, the first type of variables may be called a dormant variables, and
the second type of variable may be called a transactional variables. Each of the
different types of contracts contains both the types of variables (i.e., the first type
10 of variables and the second type of variables).
[0028] The first set of variables and the second set of variables are obtained by the
receiving module 102 for further processing. Some of the examples of the first set
of variables are provided below:
15
• Occupation (salaried or not) – salaried users were observed to have
higher complaint rate. This variable is used directly as a categorical
input (0/1)
20 • High Net-worth Indicator (HNI) status – users with high net-worth are
managed by dedicated relationship managers, and exhibit lower complaint rates. This variable is used directly as a categorical input (0/1)
25 • Distribution channel (agent, proprietary sales force, bancassurance,
online etc.) – company’s own channels such as agency and PSF exhibit lower complaint rates, while low online channels wherein there is limited customer handholding had higher complaint rates. Since there are multiple distribution channels, this variable is ‘One hot
30 encoded’
8

5
10

•

Cases where policy is expensive as indicated by higher annual premium equivalent (APE) or higher base sum assured (BSA) also exhibit higher complaint rate – this is due to higher materiality of amount associated with the policy. These continuous variables are converted into corresponding “Weight of Evidence (WoE) value.” WoE encoding transforms categorical variables into continuous variables by calculating the log odds of the target variable for each category. It helps in handling categorical data and improving model performance by providing a monotonic relationship with the target variable.

15
20

•
•

Annual income of the user – As users move from lower income buckets to higher ones, an increasing trend in the complaint rate is observed. This variable is transformed using WoE encoding
Persistency status: A higher proportion of complaints arise during the first few months of policy issuance. Once the threshold of first year is crossed, policies exhibit lower complaint rates. Therefore, persistent policies, i.e. the ones which have paid their first renewal premium tend to have lower complaint rates. This variable is used directly as a categorical input (0/1)

25
30

•

Whether medical was conducted: In case of protection policies, a cohort of users have to undergo physical or tele medical, so that insurer can assess the mortality risk before issuing the policies. Since these customers have to undergo medical assessment, and policy issuance process is more intensive than savings policies (wherein incidence of medical is lower), this cohort typically exhibits higher complaint rate. This variable is used directly as a categorical input (0/1)

9

[0029] The second set of variables employs variables relating to prior interactions
of users with various touchpoints. These variables are sourced from a user service
application that records data on every user interaction. Whenever a user contacts
5 the company via any touchpoint, say call center, branch, website etc. an interaction
ID/ call ID gets generated. This ID records aspects such as:
• Interaction date: Date on which user logged the call
• For which policy the call was logged
10 • Call source (which team did the call landed to)
• Call type – for instance was it a query or a request
• Purpose of the call – e.g. policy details updation, customer details updation, payouts, etc.
• Call owner (which team will handle the call)
15 • Committed turnaround time (TAT): For each request or query, there’s an
internal committed TAT that company follows to ensure that customers are served within appropriate timelines. These committed TATs vary with type of request or query.
• As call gets resolved, closure date is captured too
20

25

Using above fields, following variables are derived:
• Turnaround time (TAT): Calculated as closure date – interaction date. Unit: days. This variable is used to calculate average turnaround time, which is input to the model
• Average turnaround time: Calculated as average of turnaround time for all interactions pertaining to a policy. For instance, if, for a policy three interactions were recorded with TAT of 3, 1, and 3 days respectively; average TAT is calculated as (3 + 1 + 3)/3 = 2.3 days. Higher average TAT

10

is associated with higher complaint rate. This variable is transformed using one-hot encoding
• Incidence of TAT breach: Number of times TAT has exceeded committed
5 TAT. This is associated with higher complaint rate. This variable is
transformed using one-hot encoding
[0030] After the first type and the second type of variables are received by the
receiving module 102, the receiving module 102 passes on the first and the second
10 type of variables to the processing module 104 for further processing. Further, the
memory 106 stores various logistic regression models. The processing module 104 is coupled to the memory 106 and performs the processing of the received variables based on the various logistic regression models stored in the memory 106.
15 [0031] As an initial step, the univariate relationships between input variables
(various first and second set of variables) and output (complaint vs non-complaint) were quantified using information value (IV). Information Value gives a numerical value to a predictor’s ability to distinguish between different outcomes. For example, in credit scoring, it measures how well a variable can separate “goods”
20 (non-complaints) from “bads” (complaints). A higher IV indicates a stronger
predictive power of the variable. IV is calculated using following steps:
1. Divide the predictor into several groups or bins (for instance HNI vs non-
HNI can be considered two different bins)
25
2. Calculate the percentage of events (complaints) and non-events (e.g., non-
complaints) for each group.
3. Compute the Weight of Evidence (WOE) for each group, which is the
30 natural logarithm of the division of the percentage of non-events by the
percentage of events
11


4. Calculate IV as:
5 5. Variables with at least IV of 10% were used as an input to model as outlined
in next section
[0032] The receiving module 102 also obtains the third type of variables which may indicate incidence of complaint by the user. The incidence of complaint is
10 predefined for the user. The incidence of the complaint may be defined as a
frequency or rate at which a user can complaint within a specific context or over a certain period. It measures how often users express dissatisfaction or report problems with the service providers, typically in relation to products, services, or organizational processes. For instance:
15
• In a customer service setting, the incidence of complaint might refer to how
often customers contact support with issues.
• In a healthcare setting, it might pertain to how frequently patients report
20 concerns about their treatment.
[0033] Once the univariate relationships with dormant and transactional variables
were established, logistic regression (LR) models were run to understand the
multivariate relation between the first type of variables and the third type of
25 variables and the second type of variables and the third type of variables. That is
the multivariate relationship may be determined between the dormant and transactional variables with incidence of complaint. Logistic regression is a statistical method used for binary classification problems. It predicts the probability that a given input belongs to a particular category (in this complaint vs non
12

complaint). It assumes that a linear relationship between the log-odds of the dependent variable (complaint) and the independent variables (dormant/ transactional variables) as shown below:

Where: p(X) indicates probability of event (complaint in this case). X1, X2, Xp, etc. indicate input variables (dormant/ transactional models), and betas represent coefficients. A positive beta indicates that a given input variable increases the odds
10 of complaint, while a negative indicates vice versa. Model sanity is ensured by
matching sign of beta coefficients with the relation identified during the exploratory phase as well checking the statistical significance of the same. In all developed logistic regression models, variables were found to have statistically significant and intuitive coefficients.
15
[0034] The stability of the logistic regression models was ensured by splitting the data into training, testing and out of time samples. In one embodiment, model was developed on policy data for the period Mar 2019 to Sep 2022. This data was randomly split into 75:25 ratio, with model developed (or trained) on 75 pc
20 observations. Model performance was evaluated on testing set. To ensure the model
coefficients are stable,
- Model was evaluated on multiple random intime samples in the same ratio
75:25, and variability of coefficients of resultant LR equation was assessed.
25 This process is called bootstrapping. Coefficients varied in a very narrow
window which indicated model stability
13

- Model was tested on out of time period data (Oct 2022 to Mar 2023). Model performance was comparable across all three sets – training, testing and OOT
5 [0035] Using the approach detailed above, logistic regression (LR) models are
created to determine the relationship between the contract type and the first/second type of variables. For example, in one embodiment, the following 4 LR models are created:

Contract type Variables type
Protection Dormant
Protection Transactional
Savings Dormant
Savings Transactional
10 [0036] The Dormant variables represents the first type of variables, and the
Transactional variables represents the second type of variables, as explained above. Output of user who has purchased saving type policy will be a combination of the dormant variables (first type of variables) and the transactional variables (second type of variables). Output of user who has purchased protection type policy will be
15 a combination of the dormant variables (first type of variables) and the transactional
variables (second type of variables).
[0037] These LR models generate the probability of complaint as a function of
input (X) variables. For each set of X variables passed on as an input, probability
20 of complaint (Prob_Compl) is generated – Prob_Compl was observed to have very
high correlation with observed complaint rate. If one were to rank order the probability of complaint as per LR1 model in increasing order, and divide the data into 10 equal deciles, as seen in the graph below:
14

Complaints rate by deciles
4.5

1.7
0.6„_„0.9_ 1.1 ._.j|
01 0.2 0.3 04 0.5 «
1 23456789 10
Decile
• The 10th decile (the one with a tenth of observations with highest complaint
rates) have observed complaint rates that are 4.5X the complaint rate of the
entire population - probability cutoff of 10th decile was used to capture ‘R’
5 cases as per model.
• The first 5 deciles have complaints rates that are less than half of complaint
rate of entire population
10 • The complaint rates are ordinal - indicating that model is able to rank order
observations appropriately
[0038] While the LR models generate the probabilities of complaint, these are converted to ‘Red-Amber-Yellow-Green (RAYG)’ tags using heuristics. The
15 decision to classify an observation into RAYG is based on how much lift a decile is witnessing above or below the baseline. For instance, in the LR1 example used, the 10th decile that witnessed highest list (4.5X the baseline) was tagged as Red, while the next two deciles (decile 8th and 9th) which witnessed complaint rate that was 1.1X and 1.7X the baseline respectively was tagged as amber and so on.
20
[0039] By employing both first type of variables and the second type of variables (combined based on observed complaint rates in the 4*4 grid), the different contracts generates an output on RMAYG (Red Maroon Amber Yellow Green)
15

scale which indicates the user categorization based on their propensity to complain. This process too is heuristic in nature. The cross-tabulation of protection type and saving type policies, helped identify buckets of varying lift:

First type of variables→ Second type of variables ↓ R A Y G
R R (15.2X) R (7.1X) R (5.6X) M (3.3X)
A R (9.3X) M (3.2X) M (2.4X) A (1.4X)
Y A (0.9X) Y (0.5X) Y (0.4X) Y (0.2X)
G G (0.01X) G (0.01X) G (0.001X) G (0.001X)
5
[0040] In a similar manner, the savings type scheme also generates an output on
RMAYG scale that indicates the user categorization based on their propensity to
complain. These outputs are combined at customer level to generate final PTC on
RMAYG scale.
10
[0041] Models like logistic regression are used to fit the data to predict if a user is going to complain. Then the proposed invention classifies users into different tagging (Red, Maroon, Amber, Yellow, Green – i.e., RMAYG classification) based on their probability to complain. Users with Red tagging have the highest
15 propensity to complain and those with Green least. The RMAYG classification is
provided to various customer touch points (branch, call center, email, etc.) via web interface.
[0042] Referring to Figure 2 now, a flowchart of a method for categorization of
20 users based on their propensity to complain considering contract type as insurance
policy according to an embodiment of the present invention is illustrated. As explained above, the type of insurance policy can be a saving type insurance policy or a protection type insurance policy. The figure 2 has been explained considering propensity to complain of a user purchasing financial products, such as an insurance
16

policy. The various steps involved in this method can be implemented by the
logistic regression models. Particularly, each and every steps of this process can be
implemented for the protection type insurance policy as well as Savings type
insurance policy. The method begins at Step 202 where the user details and policy
5 details are extracted in the form of the first type of variables and the second type of
variables. The details can include user’s details such as occupation, income, etc., policy details such as premium, sum assured, etc. as well as history of previous customer interactions at various touch points. The first type of variables (or dormant variables) are obtained for both the protection and savings type policies. At Step
10 204, the second type of variables (e.g., transactional variables) related to previous
interaction experience indicators of users are also obtained are obtained for both the protection and savings type policies. The logistic regression models generate output on RAYG (Red, Amber, Yellow and Green) scale based on the first type of variables and the second type of variables.
15
[0043] At Step 206, RMAYG classification of users based on their propensity to complain is determined for both the saving type of policies and protection type of policies based on the dormant variables and the transactional variables. In detail, the RAYG outputs of the logistic regression model determined from both dormant
20 type of variables and transactional type of variables combined into respective
savings and protection outputs heuristically - based on the observed complaint rates of the 16 possible states (4 dormant RAYG * 4 transactional RAYG). In the 4 * 4 grid covering dormant and transactional RAYG scores, the sequential buckets with similar complaints rates are combined - the resultant 5 levels are RMAYG in
25 decreasing order of complaints rates. Protection and savings have separate merger
grids.
[0044] At Step 208, the method provides for storing the final RMAYG tagging (i.e., final RMAYG output) of users in a central database. At Step 210, the final RMAYG
17

output may be provided to the user touchpoints and targeted interventions applied based on the determined propensity to complaint of the user.
[0045] As explained above, predicting the complaining behavior of a user, and categorizing users based on their predicted probability to complain may ensures that the helpdesk team will be pre-sensitized about the type of user that they are about to interact with and ensure a better client experience. The present invention also provides the advantages of a more nuanced communication with greater focus on clarifying the aspects that figure in complaints most frequently; handling of sensitive and multiple interaction users by more experienced executives; better handholding of users during onboarding process. Some of the other advantages provided by the present invention can be regarded as:
- Inbound customer calls which select talk to agent option are routed to senior customer service executives – this allows for customers with higher PTC to be handled more empathetically
- Central tracking of open red calls allowing close coordination between multiple teams that handle customer requests/ queries
- Details of these is shared with touchpoints (e.g. call center, email, branch) and processing teams (e.g. payouts, underwriting, etc.) so that these calls are addressed on priority
- Higher preference is further given to multiple interaction customers – i.e. customers who have come back to the company with the same call type within a 90 days window
These process level interventions help enhance customer satisfaction by reducing incidence of complaints (customer satisfaction) and allow company to

utilize manpower more efficiently by allocating senior agents for customers prone to complaining
[0046] Referring to FIG. 3 now, a flowchart of a method 300 for predicting the propensity to complaint by a user is provided. At step 302, the method comprises obtaining a first type of variables. In the case of the insurance sector, the first type of variables are called dormant variables. These variables are related to the user and are obtained for both the type of insurance policies- saving type and the transaction type policies. At step 304, the method comprises obtaining a second type of variables. In the case of the insurance sector, the first type of variables are called transactional variables. These variables are related to the interaction of the user with various touch points and are obtained for both the type of insurance policies- saving type and the transaction type policies. At step 306, the method comprises obtaining a third type of variables indicating an incidence of complain by the user.
[0047] At step 308, the method comprises determining the multivariate relationship between the first type of variables and third type of variable based on first logistic regression model to predict the first probability of complaining. At step 310, the method comprises determining the multivariate relationship between the second type of variables and third type of variable based on second logistic regression model to predict the second probability of complaining. At step 312, the method comprises classifying the user based on the first probability and the second probability of complaint.
[0048] In one embodiment, the above-mentioned functions may be implemented using one or more hardware components present in a computing device. For example, the computing device may include the components as mentioned in Figure 4.
[0049] Figure 4 is an exemplary computing device 400 in which one or more embodiments of the present invention may operate, according to an embodiment.

In the system schematic of figure 4, bus 410 is in physical communication with Input/ Output device 402, interface 404, memory 406, and processor 408. Bus 410 includes a path that permits components within computing device 400 to communicate with each other. Examples of Input/ Output device 402 include peripherals and/or other mechanism that may enable a user to input information to computing device 400, including a keyboard, computer mice, buttons, touch screens, voice recognition, and biometric mechanisms and the like Input/ Output device 402 also includes a mechanism that outputs information to the user of computing device 400, such as a display, a light emitting diode (LED), a printer, a speaker, and the like.
[0050] Examples of interface 404 include mechanisms that enable computing device 400 to communicate with other computing devices and/or systems through network connections. Examples of memory 406 include random access memory (RAM), read-only memory (ROM), flash memory, and the like. The memory 406 store information and instructions for execution by processor 408. The processor 408 includes, but not limited to, a microprocessor, an application specific integrated circuit (ASIC), or a field programmable object array (FPOA) and the like. The processor 408 interprets and executes instructions retrieved from memory 406.
[0051] In one embodiment, the computing device 400 may be responsible for implementing the above-mentioned steps. For example, input parameters of the user may be received using the Input/ Output device 402. The machine learning models may be stored in the memory 406 and may be implemented by the processor 408.
[0052] In present invention, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium and executed by a hardware-based processor. The implementations described herein are not limited to any specific combinations of hardware circuitry and software.

[0053] In the drawings and specification, there have been disclosed exemplary embodiments of the invention. Although specific terms are employed, they are used in a generic and descriptive sense only and not for purposes of limitation of the scope of the invention.

WE CLAIM:
1. A method for predicting the propensity to complain by a user, the method
comprising:
obtaining a first type of variables related to the user;
obtaining a second type of variables user related to interaction with various user interaction platforms;
obtaining a third type of variables indicating incidence of the user to complain;
determining the multivariate relationship between first type of variables and third type of variable based on first logistic regression model to predict the first probability of complaining;
determining the multivariate relationship between second type of variables and third type of variable based on second logistic regression model to predict the second probability of complaining; and
classifying the user based on the first probability and the second probability of complaint.
2. The method as claimed in claim 1, wherein the first type of variables are a
dormant variables and the second type of variables are a transactional variables.
3. The method as claimed in claim 1, further comprising:
determining a univariate relationship between the first type of variables and the second type of variables.
4. The method as claimed in claim 3, wherein the output from the univariate relation
is quantified using information variable (IV);
the information variable is calculated as follows:

IV — y ( Distribution of Non-Events — Distribution of Events) X WOE all bins
wherein the WoE corresponds to the Weight of Evidence (WoE) value.
5. The method as claimed in claim 1, wherein classifying the user comprises classifying the user into Red-Amber-Yellow-Green (RAYG).
6. An apparatus for predicting the propensity to complain by a user, the apparatus comprises:
a memory;
a processor coupled to the memory and configured to perform the operations by:
obtaining a first type of variables related to the user;
obtaining a second type of variables user related to interaction with various user interaction platforms;
obtaining a third type of variables indicating incidence of the user to complain;
determining the multivariate relationship between first type of variables and third type of variable based on first logistic regression model to predict the first probability of complaining;
determining the multivariate relationship between second type of variables and third type of variable based on second logistic regression model to predict the second probability of complaining; and
classifying the user based on the first probability and the second probability of complaint.
7. The apparatus as claimed in claim 6, wherein the first type of variables are a dormant variables and the second type of variables are a transactional variables.
8. The apparatus as claimed in claim 6, further comprising:

determining a univariate relationship between the first type of variables and the second type of variables.
9. The method as claimed in claim 8, wherein the output from the univariate relation
is quantified using information variable (IV);
the information variable is calculated as follows:
IV = y (Distribution of Non-Eve nts — Distribution of Events) XWOE all bins
wherein the WoE corresponds to the Weight of Evidence (WoE) value.
10. The method as claimed in claim 6, wherein classifying the user comprises
classifying the user into Red-Amber-Yellow-Green (RAYG).