Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 10 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
Advanced Machine Learning Segmentation Model for Behavioral
Prediction in Retail Customers
William Martin Enriquez Maguiña
wenriquezm@unmsm.edu.pe
Universidad Nacional Mayor de San Marcos, Perú
Henry George Maquera Quispe
hmaquera@uncp.edu.pe
Universidad Nacional del Centro del Perú, Perú
Abstract
This study proposes an advanced segmentation model based on Machine Learning (ML) to
predict customer behavior and loyalty in the retail sector. The research adopts an applied and
quantitative approach, using two years of transactional data to optimize segmentation through
an extension of the traditional RFM model (Recency, Frequency, and Monetary value). The
methodological innovation lies in the incorporation of a new behavioral variable, derived from
predictive analysis of customer engagement, which reflects recent interaction with the brand
across multiple digital channels. This additional variable enhances the explanatory power of
the model by capturing behavioral dimensions not covered by the classic RFM framework.
Logarithmic and Box-Cox transformations were applied to correct data skewness, achieving
near-normal distributions for the original variables. The results, validated using ANOVA tests,
demonstrate statistically significant differences among the generated clusters. The enhanced
model identifies high-value customer segments more accurately and anticipates purchase and
churn patterns, resulting in more effective and personalized marketing strategies. In conclusion,
the inclusion of a complementary behavioral variable strengthens the predictive capacity of the
RFM model and reinforces its applicability as a key analytical tool for customer relationship
management in highly competitive retail environments.
Keywords: Machine learning, advanced segmentation, RFM model, behavioral prediction,
customer loyalty, retail sector.
Publicado: 17/02/2026
Aceptado: 16/02/2026
Recibido: 07/01/2026
Open Access
Article scientific
https://doi.org/10.47422/ac.v7i1.218
Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 11 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
Introduction
In today’s highly competitive retail environment,
characterized by the massive digitalization of transactions
and the proliferation of customer interaction channels,
understanding and anticipating consumer behavior has
become a strategic necessity. Companies that successfully
interpret purchasing patterns and customer motivations
gain a sustainable competitive advantage, since retaining
customers is more profitable than acquiring new ones and
contributes directly to long-term brand loyalty and
profitability [1], [2].
However, modern consumer behavior is increasingly
dynamic and complex. Digital immediacy, hyper-
personalized services, and omnichannel purchasing
experiences have generated vast amounts of heterogeneous
data that challenge traditional segmentation models [3].
The RFM (Recency, Frequency, Monetary value)
approach, widely used in marketing analytics, remains
effective for identifying high-value customers but fails to
capture behavioral, contextual, and nonlinear relationships
between variables [6], [7], [11]. Recent studies emphasize
that classical RFM-based models lack the flexibility
required to represent customer engagement and evolving
consumption patterns in digital retail environments [13].
To address these limitations, Machine Learning (ML)
provides a more robust and adaptive analytical framework.
ML algorithms can process massive datasets, uncover
hidden behavioral patterns, and predict customer actions
based on historical and contextual information [9], [12].
Specifically, ML-based segmentation models integrate
transactional, demographic, and behavioral data, allowing
the formation of more accurate and operationally
meaningful customer clusters that enhance strategic
decision-making in marketing and customer management.
The model proposed in this research introduces a new
behavioral variable into the traditional RFM framework,
extending it into an RFMC model (Recency, Frequency,
Monetary value, Category/Engagement). This additional
dimension incorporates customer engagement or product
category interaction, capturing behavioral nuances that are
not reflected in spending frequency or transaction volume.
The integration of this variable, together with
dimensionality reduction through Principal Component
Analysis (PCA), significantly increases cluster cohesion
and predictive power without compromising statistical
robustness.
From a technical standpoint, the inclusion of the
“Category” or “Engagement” variable enhances
segmentation granularity, distinguishing customers who
display similar frequency and spending patterns but differ
in product preferences or interaction levels. This
enrichment of traditional metrics expands the analytical
capacity of ML systems, allowing more precise predictions
and the delivery of personalized marketing
recommendations [8], [10].
From a commercial perspective, the implications of the
proposed model are equally relevant. By more accurately
identifying high-value segments and anticipating churn
patterns, companies can design more effective marketing
strategies, optimize resource allocation in promotional
campaigns, and strengthen loyalty programs. The synergy
between predictive analytics and commercial strategy
fosters a transformation toward data-driven marketing,
maximizing both customer satisfaction and organizational
profitability.
In summary, this research demonstrates that an advanced
Machine Learning segmentation model, enhanced with an
additional behavioral variable beyond the classical RFM
framework, provides an effective analytical tool for
predicting customer behavior and strengthening loyalty in
the retail sector. The model aligns technological innovation
with strategic business goals, contributing to
competitiveness, customer retention, and sustainable value
creation.
Literature review
2.1 Artificial Intelligence and Machine Learning in
Customer Management
Artificial Intelligence (AI) has transformed the way
companies manage customer relationships, enabling a
deeper understanding of behaviors, needs, and motivations.
Within this field, Machine Learning (ML) has emerged as
one of the most effective tools for predictive analytics and
service personalization [1], [9].
ML, as a subfield of AI, is based on algorithms that learn
from data and improve their performance without explicit
programming. Unlike traditional statistical models, ML
algorithms do not rely on rigid assumptions about data
distribution, providing greater flexibility to detect
nonlinear and hidden patterns among variables [9]. This
adaptability is especially valuable in the retail sector, where
information is massive, heterogeneous, and continuously
evolving.
According to Kühl et al. [9], ML-driven service systems
have evolved from descriptive models toward adaptive
systems that integrate transactional, demographic, and
contextual data. Similarly, Aguiar-Costa et al. [1]
demonstrated that AI-driven customer service applications
significantly increase satisfaction and loyalty by enabling
Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 12 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
faster, more personalized, and consistent responses
throughout the customer lifecycle.
2.2. Customer Behavior Analysis and Segmentation
Models
Customer behavior analysis constitutes the foundation of
strategic management in the retail industry. Customer
segmentation enables the classification of consumers with
similar behavioral and purchasing characteristics,
facilitating the design of more precise and efficient
marketing strategies [6], [7].
Traditionally, the RFM (Recency, Frequency, Monetary
value) model has been one of the most widely used
approaches for this purpose. Originally introduced by
Hughes (1994) and later refined by Cheng and Chen [6],
this method evaluates customer value and loyalty through
three key metrics:
• Recency (R): Time elapsed since the last purchase.
• Frequency (F): Number of purchases during a defined
period.
• Monetary value (M): Total spending within that
period.
Although simple and practical, the classical RFM model
presents significant analytical limitations. It excludes
behavioral and contextual dimensions, such as digital
engagement or product category preferences, which are
crucial in understanding modern consumers [4], [7], [11].
Moreover, its reliance on static rules limits its adaptability
to highly dynamic digital environments, reducing its long-
term effectiveness [3], [13].
To overcome these limitations, Machine Learning-based
segmentation models have demonstrated higher predictive
accuracy and adaptability. These models combine
unsupervised clustering methods, such as K-means, with
supervised classification or regression algorithms to
identify loyalty and churn patterns [5], [8]. According to
Aldunate et al. [3], deep learning architectures can enhance
segmentation by analyzing customer satisfaction through
natural language processing, allowing the inclusion of new
behavioral dimensions in predictive modeling.
2.3. Extending the RFM Model through Behavioral
Variables
In recent years, several studies have proposed extensions
of the traditional RFM framework by incorporating
additional behavioral dimensions. Recent research
highlights that variables related to customer engagement,
interaction intensity, or product category preference
significantly improve segmentation accuracy and
predictive performance [11], [12], [15].
These enriched models allow a more granular
interpretation of customer value and loyalty, particularly in
data-rich retail contexts. Hu and Yeh [7] introduced the
RFMP model (Recency, Frequency, Monetary, Pattern),
which integrates purchase sequence and temporal
frequency, while Das and Nayak (2022) and Al-Araj et al.
[2] suggested the inclusion of digital interaction metrics to
enhance loyalty prediction.
This research introduces an RFMC model, which adds a
behavioral variable named Category or Engagement to the
traditional framework. This variable measures the level of
customer interaction with the brand or specific product
categories, expanding the analytical perspective of the
RFM model. It captures qualitative dimensions of customer
behavior —such as preferences, product types, or
emotional connection to the brand— that are essential for
predicting long-term loyalty and value [10].
From a technical standpoint, the additional variable
enhances cluster cohesion and discrimination, allowing for
the separation of customer segments with similar spending
and frequency patterns but different behavioral profiles.
Kumar and Shah [10] argue that early identification of such
high-value behavioral segments enables the development
of targeted retention strategies and increases Customer
Lifetime Value (CLV).
2.4. Commercial Impact of Predictive Analytics in
Retail
The application of predictive analytics in the retail industry
represents not only a technological advancement but also a
major commercial transformation. Organizations that
adopt data-driven marketing strategies improve campaign
efficiency, reduce acquisition costs, and strengthen
customer loyalty [2], [8].
Kietzmann et al. [8] emphasize that AI-driven marketing
enables real-time decision automation and personalized
customer experiences, which increase satisfaction and
perceived value. Similarly, Al-Araj et al. [2] demonstrate
that AI integration in service quality management enhances
consumer trust by aligning responses with customer
expectations.
In this context, the proposed advanced segmentation model
based on ML with an additional behavioral variable not
only increases analytical precision but also reinforces the
commercial competitiveness of retail companies. By
anticipating customer behavior and identifying high-value
segments, firms can optimize marketing resources, design
personalized loyalty programs, and achieve sustainable
Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 13 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
market advantages that translate into improved
profitability indicators [5], [10].
Methodology
3.1 Research Approach
This research follows an applied, quantitative, and
explanatory approach, aimed at developing a predictive
customer segmentation model based on Machine Learning
(ML). Its primary objective was to enhance the accuracy of
customer behavior identification and loyalty prediction
within the retail sector.
The study was structured into three main phases:
• Data preprocessing and preparation.
• Construction and optimization of the advanced
segmentation model (RFMC).
• Statistical evaluation and predictive validation.
Each phase was designed to ensure data consistency, model
robustness, and reliability of results, combining techniques
from data mining, statistical analysis, and machine
learning.
3.2. Data Source and Description
The dataset was obtained from the transactional
information system of a Peruvian retail company operating
in the fast-moving consumer goods sector. It included
customer records covering a two-year period, comprising
more than 120,000 individual transactions.
Each record contained the following information:
• Transaction date and monetary amount (temporal and
financial dimensions).
• Customer purchase frequency over monthly and
quarterly intervals.
• Product category purchased (new attribute
incorporated).
• An anonymized customer identifier to preserve data
privacy.
The original data were transformed into RFM behavioral
metrics:
• Recency (R): Number of days since the last purchase.
• Frequency (F): Number of purchases made during the
analysis period.
• Monetary Value (M): Total spending by the customer.
• Category / Engagement (C): Degree of interaction or
recurrence within a specific product category.
The inclusion of the C (Category/Engagement) variable
represents the innovative contribution of the model, as it
introduces a behavioral dimension that expands the
explanatory power of traditional metrics [2], [10].
3.3. Data Preprocessing and Variable Transformation
Prior to modeling, the data underwent a comprehensive
cleaning, standardization, and transformation process to
ensure analytical quality and consistency. The following
stages were implemented:
• Outlier and duplicate removal.
• Natural logarithmic transformation for highly skewed
variables (R, F, M).
• Box-Cox transformation to improve distribution
normality and stabilize variance.
• Min-max scaling to standardize variable magnitudes
before applying unsupervised learning algorithms.
These transformations reduced distortion caused by
extreme customer behaviors and improved cluster
cohesion—an essential step for enhancing clustering
quality [4], [7].
3.4. RFMC Model Design
The proposed model was built upon the traditional RFM
framework, enhanced by adding the C
(Category/Engagement) variable, giving rise to the
RFMC model. The objective was to enrich customer
segmentation by integrating qualitative behavioral
information.
A combination of unsupervised and supervised ML
techniques was applied:
• Initial segmentation:
The K-means algorithm was used to group customers
according to similarities in their RFMC profiles. The
optimal number of clusters was determined through the
elbow method and silhouette coefficient, ensuring a
balance between intra-cluster cohesion and inter-cluster
separation.
• Dimensionality reduction:
Principal Component Analysis (PCA) was applied to
evaluate the variance explained by the four dimensions of
the model. The first two components accounted for over
85% of the total variance, demonstrating model
consistency.
Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 14 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
• Predictive classification:
A multinomial logistic regression model was trained to
predict the probability of new customers belonging to each
cluster. This integration of supervised and unsupervised
learning produced a hybrid ML architecture with high
predictive performance [6], [9].
3.5. Statistical Model Validation
Model validation was conducted at two levels:
• Internal statistical validation using ANOVA tests at a
95% confidence level. Results showed statistically
significant differences (p < 0.05) among the generated
clusters, confirming the consistency of the
segmentation.
• External and predictive validation using a test sample
comprising 30% of the total dataset. The overall
classification accuracy reached 87.4%, outperforming
the traditional RFM model and demonstrating a
substantial improvement in behavioral prediction
capability.
These results support the central hypothesis of the study:
the inclusion of an additional behavioral variable increases
the robustness of the model and its practical applicability
for customer loyalty management.
3.6. Tools and Development Environment
Data processing and modeling were conducted in a
reproducible analytical environment using Python (version
3.10) and specialized libraries such as Pandas, Scikit-learn,
NumPy, and Matplotlib.
Additionally, Power BI and Tableau were used for data
visualization and representation of segmentation outcomes
and performance indicators.
This technological ecosystem enabled the integration of
data mining, exploratory analysis, machine learning, and
interactive visualization techniques, ensuring a robust and
replicable methodological workflow.
3.7. Ethical and Confidentiality Considerations
All analyzed data were anonymized to ensure customer
privacy. The research adhered to ethical standards and data
protection regulations in accordance with Peruvian Law
No. 29733 (Personal Data Protection Act) and
international best practices in applied research ethics.
Results and Discussion
4.1. RFMC Model Results
The results confirm the validity of the RFMC model as a
significant improvement over the classical RFM
framework. The inclusion of the new variable C
(Category/Engagement) allowed the model to capture
behavioral dimensions that traditional metrics could not,
thereby enhancing segmentation accuracy.
Fig. 1. Visual comparison between RFM and RFMC
customer segmentation models.
Source: Author’s own elaboration
After applying logarithmic and Box-Cox transformations,
the distributions of the Recency, Frequency, and Monetary
Value variables approached statistical normality. This
correction reduced data dispersion and improved the
model’s sensitivity to outliers, reinforcing its clustering
performance.
Fig. 2. Bivariate Distribution Comparison: RFM vs.
Enhanced RFMC
Source: Author’s own elaboration
Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 15 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
The K-means algorithm, optimized through cohesion and
separation metrics (silhouette = 0.82), generated five well-
differentiated clusters with clearly identifiable behavioral
profiles. The ANOVA test revealed statistically significant
differences among these groups (p < 0.05), validating the
robustness of the model in distinguishing purchasing
behavior patterns.
4.2. Cluster Analysis and Behavioral Patterns
The category concentration analysis demonstrated that the
clusters generated by the RFMC model exhibited higher
internal homogeneity and greater coherence with product
preferences compared to the traditional RFM model. Key
findings include:
• Cluster 1 showed strong affinity for Gifts and
Decoration (19%) and Fashion and Jewelry (17%),
representing clients with well-defined preferences and
purchase motivation driven by special occasions.
• Cluster 3, more diversified, was characterized by
interest in Gifts and Decoration (16%) and other
categories (15%), suggesting emotional or event-
driven purchase behavior.
• Clusters 0, 2, and 4 exhibited more balanced category
distributions, corresponding to generalist or
exploratory customers—ideal targets for loyalty-
building strategies.
These patterns were visualized through a heatmap of
average category distribution per cluster, which clearly
reflected preference concentration and validated the
model’s effectiveness in capturing the underlying logic of
customer behavior [11].
4.3. Statistical Validation and Hypothesis Testing
The ANOVA test results confirmed statistically significant
differences across clusters for the variables Price and
Quantity (p < 0.05). Consequently, the null hypothesis (H₀)
was rejected and the alternative hypothesis (H₁) accepted:
“The RFMC model significantly improves the
effectiveness and accuracy of customer preference
identification in a retail company.”
Additionally, cross-validation with a 30% test sample
yielded an overall classification accuracy of 87.4%,
substantially outperforming the traditional RFM model.
This empirical evidence supports the study’s central
hypothesis: the inclusion of an additional behavioral
variable (C) enhances both the explanatory power and
predictive capacity of the segmentation model.
4.4. Discussion and Commercial Implications
From a commercial perspective, the findings reveal
strategic implications for retail customer management.
The RFMC model enables:
• Identification of high-value customers and prediction
of repurchase or churn likelihood, optimizing
marketing resource allocation.
• Design of personalized campaigns based on the
predominant category of each cluster, increasing
conversion rates and return on investment.
• Implementation of dynamic loyalty programs, adjusted
to real customer behavior rather than historical
spending alone.
Fig. 3. Average product category distribution by RFMC
cluster.
Source: Author’s own elaboration
These results align with recent studies on artificial
intelligence in predictive marketing (Aguiar-Costa et al.,
2022; Sun et al., 2019), which highlight the capacity of ML
algorithms to enable deeper segmentation and detect
emerging behavioral trends in competitive environments.
Therefore, the RFMC model demonstrates not only
technical superiority over the traditional RFM approach but
also tangible commercial value, offering a more
comprehensive view of customer behavior and supporting
data-driven decision-making in highly dynamic retail
markets.
These findings are consistent with prior research indicating
that ML-based customer segmentation models significantly
outperform traditional rule-based approaches in both
predictive accuracy and managerial relevance [12], [14].
Furthermore, the integration of behavioral variables into
segmentation frameworks has been shown to strengthen
Revista de Investigación Científica y Tecnológica
Alpha Centauri
ISSNe: 2709-4502
PROF-AC-V-00
Vol. 7 N.º 1
(Enero - Marzo, 2026)
Página 16 de 16
Revista de Investigación Científica y Tecnológica Alpha Centauri
customer loyalty strategies by enabling personalized
interventions and long-term value optimization [15].
References
[1] L. M. Aguiar-Costa, C. A. X. C. Cunha, W. K. M. Silva,
and N. R. Abreu, “Customer satisfaction in service delivery
with artificial intelligence: A meta-analytic study,” Revista
de Administração Mackenzie, vol. 23, no. 6, 2022.
[2] R. Al-Araj, H. Haddad, M. Shehadeh, E. Hasan, and M.
Y. Nawaiseh, “The effect of artificial intelligence on
service quality and customer satisfaction in the Jordanian
banking sector,” WSEAS Transactions on Business and
Economics, vol. 19, pp. 1929–1947, 2022.
[3] Á. Aldunate, S. Maldonado, C. Vairetti, and G.
Armelini, “Understanding customer satisfaction via deep
learning and natural language processing,” Expert Systems
with Applications, vol. 209, 118309, 2022, doi:
10.1016/j.eswa.2022.118309.
[4] A. Hiziroglu and S. Sengul, “Customer segmentation
using data mining techniques in retail banking,”
International Journal of Business and Social Science, vol.
3, no. 24, pp. 252–262, 2012.
[5] S. Rosset, “Modeling customer lifetime value,” Journal
of Marketing Research, vol. 40, no. 3, pp. 321–339, 2003.
[6] C.-H. Cheng and Y.-S. Chen, “Classifying the
segmentation of customer value using RFMP variables and
cluster analysis,” Expert Systems with Applications, vol.
36, no. 3, pp. 4176–4184, 2009, doi:
10.1016/j.eswa.2008.04.003.
[7] Y.-H. Hu and I.-C. Yeh, “Discovering valuable frequent
patterns based on RFM analysis without customer
identification information,” Knowledge-Based Systems,
vol. 61, pp. 76–88, 2014, doi:
10.1016/j.knosys.2014.02.009.
[8] J. Kietzmann, J. Paschen, and E. Treen, “Artificial
intelligence in advertising: How marketers can leverage AI
and machine learning,” International Journal of
Advertising, vol. 37, no. 3, pp. 333–347, 2018.
[9] N. Kühl, M. Goutier, G. Satzger, and B. Niehaves,
“Machine learning in service systems: A systematic
literature review,” Journal of Business Research, vol. 139,
pp. 1313–1330, 2022.
[10] V. Kumar and D. Shah, “Building and sustaining
profitable customer loyalty for the 21st century,” Journal
of Retailing, vol. 80, no. 4, pp. 317–330, 2004.
[11] T. Ho, S. Nguyen, H. Nguyen, N. Nguyen, D.-S. Man,
and T.-G. Le, “An extended RFM model for customer
behaviour and demographic analysis in the retail industry,”
Business Systems Research Journal, vol. 14, no. 1, pp. 26–
53, 2023, doi: 10.2478/bsrj-2023-0002.
[12] S. Das and J. Nayak, “Customer segmentation via data
mining techniques: State-of-the-art review,” in
Computational Intelligence in Data Mining, Smart
Innovation, pp. 489–507, 2022, doi: 10.1007/978-981-16-
9447-9_38.
[13] A. Griva, C. Bardaki, K. Pramatari, and G. Doukidis,
“Factors affecting customer analytics: Evidence from three
retail cases,” Information Systems Frontiers, vol. 24, no. 2,
pp. 493–516, 2022, doi: 10.1007/s10796-020-10098-1.
[14] Y. Sun, S. Wang, and M. Zhang, “Machine learning–
based customer personalization and loyalty management in
retail,” Journal of Retailing and Consumer Services, vol.
50, pp. 102–112, 2019. (Referencia consignada en la tesis)
[15] A. S. Dick and K. Basu, “Customer loyalty: Toward
an integrated conceptual framework,” Journal of the
Academy of Marketing Science, vol. 22, no. 2, pp. 99–113,
1994, doi: 10.1177/0092070394222001.
