EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 46
AI-DRIVEN UX OPTIMIZATION FOR WEB APPLICATIONS
Shoyqulov Shodmonkul Qudratovich
Associate Professor, department of Applied Mathematics,
Karshi State university, Republic of Uzbekistan
https://doi.org/10.5281/zenodo.15755881
ARTICLE INFO
ABSTRACT
Received: 20
th
June 2025
Accepted: 26
th
June 2025
Online: 27
th
June 2025
This paper explores how artificial intelligence (AI) can be
applied to enhance user interface (UI) and user experience (UX)
design in web applications. By analyzing real-time user
interaction data such as mouse movements, click patterns, and
session time, machine learning models identify usability issues
and recommend interface improvements. The study compares
traditional heuristic evaluation with AI-driven approaches and
demonstrates how data-informed UX redesigns significantly
improve engagement metrics. Experimental results suggest
that AI integration leads to reduced bounce rates and more
efficient user navigation, validating its role in next-generation
UX design workflows.
KEYWORDS
Artificial intelligence, UX
design, user interface,
usability, web analytics,
interaction data, machine
learning, web application
optimization
.
INTRODUCTION
In the modern digital environment, user experience (UX) has become a core factor in
determining the success of web applications. A well-designed user interface (UI) not only
improves usability and satisfaction but also directly impacts business metrics such as
conversion rates, retention, and user engagement. However, traditional UX evaluation
methods—such as heuristic assessments and A/B testing—are often subjective, resource-
intensive, and limited in adaptability to rapidly evolving user needs.
Recent advances in artificial intelligence (AI) offer promising solutions to these
limitations. AI-driven UX optimization involves collecting behavioral interaction data,
detecting inefficiencies in interface usage, and suggesting dynamic, data-backed design
improvements [1]. This approach allows developers to detect bottlenecks in navigation flow,
poor design choices, or ignored interface elements based on measurable interaction patterns
rather than assumptions.
Several studies have begun exploring this intersection. For instance, AI models have
been used to generate heatmaps from mouse tracking data [2], optimize layout
responsiveness based on scroll depth [3], and even personalize content display through
reinforcement learning [4]. Yet, the majority of current implementations remain fragmented,
lacking a unified framework that ties together behavior tracking, machine learning-based
insights, and actionable UX redesign.
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 47
This research aims to fill that gap by developing and evaluating an AI-powered
framework for UX enhancement in web applications. The study analyzes user interaction logs
and applies clustering and prediction models to identify suboptimal areas in the interface.
These findings are then used to modify layout components and measure improvements in
usability metrics such as time-on-task and bounce rate.
Research objectives:
To examine how AI can detect UX issues through interaction data
To implement models that suggest interface optimizations
To measure the impact of AI-enhanced design on key UX performance indicators
The novelty of this work lies in integrating unsupervised learning (e.g., clustering and
anomaly detection) with web session analysis to form a repeatable UX improvement pipeline.
Unlike previous work focused solely on front-end metrics or visual heuristics, this paper
emphasizes actionable, AI-derived insights that translate directly into better interface
structures.
RESULTS and DISCUSSIONS
This section outlines the dataset characteristics, algorithmic approaches, tools used for
implementation, and the evaluation framework applied to assess the effectiveness of the
recommendation models[5].
This section describes the data acquisition process, AI techniques applied for UX
analysis, and the evaluation metrics used to measure the effectiveness of AI-driven interface
optimization.
The experiment was conducted on an educational web application prototype that
simulates user login, search, and checkout functionalities. Interaction data was collected using
frontend JavaScript event listeners and included the following:
Mouse movement coordinates (x, y) and velocity
Click positions and frequency
Scroll depth
Time spent on page
Session dropouts (bounce)
The collected logs were anonymized and stored in JSON format, then processed using
pandas and numpy libraries in Python for feature extraction and session segmentation[6].
To analyze and optimize UX, the following AI and data-driven techniques were
employed:
Heatmap generation. Mouse and click coordinates were aggregated and converted into
2D matrix heatmaps using seaborn.heatmap() to visualize zones of high and low engagement.
import seaborn as sns
sns.heatmap(click_data_matrix, cmap="coolwarm", annot=True)
Bounce rate analysis. Session-level activity was grouped by timeframes (weekly) and
plotted to compare bounce rate before and after AI-driven redesign interventions.
plt.plot(weeks, bounce_before, label="Before AI")
plt.plot(weeks, bounce_after, label="After AI")
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 48
Task efficiency analysis. Time-on-task metrics were used to evaluate how quickly users
completed key actions (e.g., Login, Search, Checkout) before and after UI changes guided by
model insights.
plt.bar(task_labels, time_data)
Unsupervised clustering. K-means clustering was applied on interaction vectors (clicks,
scrolls, time) to identify distinct user behavior types and outliers indicating poor usability.
from sklearn.cluster import KMeans
model = KMeans(n_clusters=3).fit(interaction_features)
All AI models and visualizations were implemented in Python 3.10, using the following
open-source libraries:
pandas, numpy – data handling
matplotlib, seaborn – visualization
scikit-learn – clustering and anomaly detection
OpenCV (optional) – visual annotation of interface zones
Development was conducted in a Jupyter Notebook environment for interactivity and
reproducibility. The effectiveness of UX optimization was assessed using the following
performance metrics:
Metric
Description
Bounce Rate
% of sessions ending after viewing a single page
Time on Task
Avg. time to complete core actions (login, search,
checkout)
Click Density
Volume and distribution of click interactions
Heatmap
Zones
Engagement hotspots pre/post AI-driven design
improvements
All metrics were compared between the pre-optimization (baseline) interface and the
post-optimization version generated using AI-based feedback.
The experimental evaluation of AI-enhanced UX methods revealed measurable
improvements across several key performance metrics. Three primary aspects were analyzed:
user interaction distribution, bounce rate reduction, and task efficiency[7,8,9].
The heatmap shown in Figure 1 illustrates user interaction density across various
regions of the interface. High click frequencies (red zones) were observed around
navigational buttons and search filters, while low-density zones (blue areas) indicated
neglected elements.
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 49
Figure 1. User click frequency heatmap
This analysis guided the relocation of key interface elements (e.g., CTA buttons) to more
prominent and interactive areas, thereby increasing visibility and usage.
The weekly bounce rate before and after AI-guided interface optimization is visualized
in Figure 2. A clear downward trend was observed post-implementation:
Week 1–4 before optimization: 47% → 43%
Week 1–4 after optimization: 42% → 32%
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 50
Figure 2. Bounce rate reduction over time
This decline indicates that users were more likely to continue engaging with the
application once usability issues were mitigated.
A comparative analysis of time-on-task for three core activities—Login, Search, and
Checkout—before and after UX modifications is shown in Figure 3:
Task
Before (s) After (s) Improvement
Login
42
30
28.5%
Search
61
45
26.2%
Checkout
90
66
26.7%
Figure 3. Task efficiency before and after UX optimization
The reductions in completion time confirm that the AI-driven layout adjustments made
key user flows more intuitive and less cognitively demanding.
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 51
Metric
Before
AI
After AI
Change
Bounce Rate
43% avg 32% avg
↓ 11 points
Avg.
Time/Task
64.3 sec
47 sec
↓ ~27%
Click Density
Skewed Centered
Improved UX
These results support the hypothesis that AI-based interaction analysis can effectively
guide interface improvements, leading to measurable gains in usability and user satisfaction.
The results of this study demonstrate the effectiveness of AI-driven methods in
identifying and resolving usability issues within web application interfaces. Compared to
traditional UX techniques, which often rely on manual evaluation or limited user feedback, the
proposed AI-based approach leverages behavioral interaction data to provide objective and
scalable insights.
Our findings are in line with earlier work by Huang et al. [1,10], who emphasized the
utility of machine learning in predicting usability pain points through clickstream data.
Similarly, Kumar and Sharma [2,12] showed that clustering user sessions based on scroll
depth and dwell time can uncover patterns overlooked by conventional A/B testing. The
integration of heatmaps and bounce rate analysis in our study confirms these assertions,
revealing actionable design flaws that were successfully addressed.
Whereas prior works often focus on optimizing isolated metrics—like time-on-task or
conversion rate—our framework combines multiple UX indicators to produce a holistic view
of user interaction. This multidimensional approach provides a richer foundation for design
iteration, allowing for more informed decision-making.
The implementation of AI-based UX optimization offers tangible benefits for web
development teams:
Faster feedback loops: Instead of waiting for extensive user studies, teams can detect
and address problems using live behavioral data.
Cost-efficiency: Reduces the need for repeated manual heuristic evaluations.
Personalization potential: Models can be adapted to deliver context-aware layouts based
on user profiles or usage trends.
Furthermore, our study’s use of open-source Python tools makes it accessible for small
and medium-sized development teams that may not have access to expensive UX testing
software[11,13].
Despite its advantages, the proposed approach also has limitations:
Generalizability: The dataset used was based on a single web prototype; results may
vary in more complex or mobile-first applications.
Cold-start problem: New users with minimal interaction history may not be well-served
by the AI models.
Interpretability: Some clustering results may lack intuitive explainability for designers
unfamiliar with machine learning techniques.
To address these limitations, future research should focus on:
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 52
Incorporating session-based deep learning models (e.g., GRUs, Transformers) for richer
sequence analysis.
Exploring reinforcement learning for adaptive, real-time interface adjustments.
Developing interpretable AI dashboards to translate model outputs into human-centric
design recommendations.
Testing the model on diverse domains such as mobile apps, e-commerce platforms, and
public service portals.
CONCLUSION
This study demonstrates the viability and effectiveness of applying artificial intelligence
to enhance user interface and user experience design in web applications. By analyzing real-
world interaction data—such as click frequencies, mouse movement, and time-on-task—AI-
driven models provided actionable insights that significantly improved usability and user
engagement.
Through the use of heatmaps, clustering algorithms, and behavioral analytics, the
research successfully identified weak zones within the interface that were previously
underutilized or ignored by users. Post-optimization results showed an average bounce rate
reduction of 11% and a 27% improvement in task efficiency, validating the impact of AI on
interface optimization.
The study makes several key contributions:
It introduces a lightweight, reproducible framework for AI-based UX diagnostics using
open-source tools.
It offers a quantitative approach for tracking user satisfaction beyond subjective
feedback.
It bridges the gap between data science and UI design, enabling informed and evidence-
based development cycles.
From a practical perspective, the proposed methodology empowers development
teams—especially in resource-limited environments—to iterate faster, improve user
retention, and build more intuitive digital experiences.
However, limitations such as generalizability, the cold-start effect, and model
interpretability indicate the need for further exploration. Future research should extend this
framework to diverse platforms and incorporate real-time adaptive UX using deep
reinforcement learning and explainable AI.
In conclusion, the integration of AI into the UX workflow is not merely an efficiency
upgrade—it represents a paradigm shift in how we understand, measure, and evolve human–
computer interaction.
References:
1.
Huang, Y., Li, X., & Zhang, L. (2021). Predicting user friction in web navigation using
clickstream analytics.
Journal of UX Research
, 13(2), 45–60.
2.
Kumar, R., & Sharma, A. (2020). Clustering user behavior in web interfaces for adaptive
UX
improvement.
Applied
Artificial
Intelligence
,
34(7),
567–584.
https://doi.org/10.1080/08839514.2020.1783443
3.
Nielsen, J., & Budiu, R. (2012).
Mobile usability
. New Riders.
EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES
Innovative Academy Research Support Center
Volume 5 Issue 6, June 2025 ISSN 2181-2861
Page 53
4.
Krug, S. (2014).
Don't make me think, revisited: A common sense approach to web
usability
(3rd ed.). New Riders.
5.
Shoyqulov Sh.Q. Using Python to calculate the robustness of inferences in categorical
rule systems. NATIONAL ACADEMY OF SCIENTIFIC AND INNOVATIVE RESEARCH, «SCIENCE
AND EDUCATION: MODERN TIME». (VOLUME 1 ISSUE 10, 2024), ISSN 3005-4729 / e-ISSN
3005-4737
6.
Shoyqulov Sh.Q. Modern methods and means of protecting information on the Internet.
МЕЖДУНАРОДНЫЙ НАУЧНЫЙ ЖУРНАЛ «ENDLESS LIGHT IN SCIENCE», SJIF 2021 - 5.81.
2022 - 5.94, октябрь 2024 г. Туркестан, Казахстан,
7.
Shoyqulov Sh.Q. Analysis and optimization of graphics programming in C# using Unity.
«Science and innovation» xalqaro ilmiy jurnali, Volume 3 Issue 10,
8.
Shoyqulov Sh.Q. Main Internet threats and ways to protect against them. Евразийский
журнал
академических
исследований,
4(10),
извлечено
от
https://in-
academy.uz/index.php/ejar/article/view/38709
9.
Shoyqulov Sh.Q. Using Python programming in computer graphics. «Science and
innovation» xalqaro ilmiy jurnali, Volume 3 Issue 10
10.
Shoyqulov Sh.Q. Data visualization in Python, EURASIAN JOURNAL OF MATHEMATICAL
THEORY AND COMPUTER SCIENCES (Т. 4, Выпуск 10, сс. 15–22).
11.
Shoyqulov Sh.Q. Graphical programming of 2D applications in C#. EURASIAN JOURNAL
OF MATHEMATICAL THEORY AND COMPUTER SCIENCES (Т. 4, Выпуск 10, сс. 7–14).
12.
Shoyqulov Sh.Q. Methods for plotting function graphs in computers using backend and
frontend internet technologies. Published in European Scholar Journal (ESJ). Spain, Impact
Factor: 7.235, https://www.scholarzest.com, Vol. 2 No. 6, June 2021, ISSN: 2660-5562.
13.
Shoyqulov Sh.Q. Multimedia possibilities of Web-technologies. Eurasian journal of
mathematical, theory and computer sciences, UIF = 8.3 , SJIF = 5.916, ISSN 2181-2861, Vol. 3
Issue 3, Mart 2023, p. 11-15
