Impact of Smartphone Usage on Students’ Academic Performance Using Contemporary Deep Learning Model
S. Vimala1, G. Arokia Sahaya Sheela2
International Journal of Information Technology, Research. & Applications ISSN: 2583-5343, Vol. 4 No. 4, Dec 2025 (01-08)
1,2Department of Computer Science, St. Joseph’s College(Autonomous), Tiruchirappalli -2, Affiliated to Bharathidasan University, Tamil Nadu, India.
Article history:
Received Sep 30, 2025
Revised Nov 15, 2025
Accepted Now 30, 2025
Keywords:
Digital Learning Analytics
Behavioral Pattern Recognition
Academic Performance Prediction
Hybrid Neural Architectures
Educational Technology Impact
ABSTRACT
The rapid growth of smartphone usage among students has created both opportunities and challenges in the academic environment. The goal of this study is to examine how different smartphone usage behaviors—such as screen time, application preferences, and late-night activity—affect student academic performance. The methods combine advanced deep learning approaches, including Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM), with carefully preprocessed smartphone interaction data collected from university students between 2023 and 2025. The models were trained to capture both spatial and sequential usage patterns, with hybrid architectures applied to maximize predictive capability. The findings show that deep learning models consistently outperform traditional machine learning techniques, with the CNN-BiLSTM hybrid achieving the highest accuracy of 92.4%. This confirms that smartphone usage patterns are strong predictors of academic outcomes, particularly when late-night use and excessive social media activity are present. The novelty of this research lies in its integration of diverse behavioral features—ranging from app-specific time allocation to multitasking frequency—into deep learning pipelines, offering actionable insights for educators and policymakers. This work advances the field of educational data science by providing a reliable framework for predicting performance and guiding balanced digital practices among students.
This is an open access article under the CC BY-SA license
Corresponding Author:
Vimala S
Department of Computer Science,
St. Joseph’s College (Autonomous),
Tiruchirappalli -02,
Affiliated to Bharathidasan University,
Tamil Nadu, India.
Email: vimalas_phdcs@mail.sjctni.edu

1 Introduction

The ubiquity of smartphones has fundamentally reshaped how students communicate [9], access information, and manage their daily routines. While smartphones offer vast educational resources, their excessive and unregulated use has become a double-edged sword, leading to potential distractions, reduced focus, and sleep deprivation. Consequently, understanding the relationship between smartphone usage and academic performance has emerged as a crucial area of academic inquiry [10, 11].
Educational researchers have observed that students spend a significant portion of their time on social networking applications, gaming platforms, and streaming services [12], often at the expense of study-related activities. This behavioral shift influences not only cognitive performance but also emotional and psychological well-being. With global concerns regarding declining attention spans, it becomes imperative to investigate how digital behavior correlates with student outcomes [13].
Recent advancements in deep learning technologies have revolutionized predictive analytics in education. Unlike traditional machine learning methods that rely heavily on manual feature selection, deep learning models autonomously extract meaningful representations from complex and unstructured data [14, 15]. For instance, CNNs effectively capture spatial correlations in app usage frequencies, while recurrent architectures like Bi-LSTM can identify sequential behavior such as late-night browsing or repeated gaming sessions before exams.
The primary objective of this study is to employ state-of-the-art deep learning models to analyze patterns in smartphone usage and predict their impact on academic performance [17]. We focus on high school and university-level students between 2023 and 2025, utilizing large-scale datasets collected through both self-reported surveys and passive digital tracking. By integrating behavioral features [19] such as time spent per category, multitasking frequency, and irregular study schedules, we provide a comprehensive framework for academic performance prediction.
This research aims to advance three critical contributions:
  1. Efficient preprocessing techniques to handle raw smartphone interaction logs.
  2. Feature engineering and deep learning integration to identify the most relevant behavioral attributes.
  3. Interpretability of predictive outcomes, offering insights for educators and parents to develop balanced usage strategies.
By bridging digital behavioral analytics with cutting-edge computational intelligence, we position this study as a pivotal contribution to the emerging field of educational data science [20].

2 Literature Review

Recent research has increasingly explored the relationship between smartphone usage and academic outcomes, with deep learning models offering fresh insights. A review of recent studies (2022–2025) reveals both consistent findings and new methodological directions.
Vimala S et al. [1] stressed the importance of data preprocessing and attribute selection in reducing computational complexity. They argued that careful feature engineering not only saves processing time but also helps models maintain accuracy when applied to large-scale educational datasets. Ellikkal et al. [2] introduced deep reinforcement learning techniques for personalized study reminders. Their system adaptively adjusted notification timing based on prior usage, which helped counter excessive device use and nudged students toward healthier study patterns. Rizwan et al. [3] emphasized app-switching frequency as a strong predictor of learner distraction. Their results also indicated that frequent toggling between unrelated apps is a reliable marker of fragmented attention, which directly undermines sustained learning efforts.
Raj et al. [4] observed a direct connection between excessive gaming and reduced assignment completion rates. They added that prolonged gaming not only consumes study time but also diminishes motivation for academic tasks, highlighting the need for balancing digital entertainment with study routines. Hemal et al. [5] incorporated emotional quotient (EQ) indicators alongside smartphone activity logs, enhancing interpretability and practical relevance of predictive models. Their approach suggested that emotional resilience may buffer some of the negative consequences of high device use, providing a more holistic view of learner performance.
Xu, Z . [6] investigated the influence of late-night smartphone activity, linking it to disrupted sleep cycles and lower GPAs. Their study further suggested that irregular sleep patterns decrease concentration and memory retention, amplifying the negative impact on learning efficiency. Hemdanou et al. [7] compared deep learning models with ensemble approaches, finding Bi-LSTM especially effective in detecting sequential interruptions during study sessions. The authors also underlined that sequence-aware models provide richer insights into behavioral shifts, particularly when students frequently pause or switch tasks.
Al-Alawi et al.[8]proposed a hybrid CNN-BiLSTM framework, which achieved accuracy levels above 90% when predicting student results from raw usage logs. They also showed that the combination of spatial and temporal feature extraction allows for more robust modeling of student behaviors compared to using either method alone. Hussain, S et al.[16] highlighted that convolutional neural networks (CNNs) are more effective than Random Forest models in identifying high-risk usage behaviors. Their work also stressed that CNNs capture subtle time-based usage variations that traditional models often overlook, making them valuable for early detection of problematic patterns. Zheng, L et al. [18] demonstrated that multitasking between learning apps and entertainment platforms—such as social media or video streaming—negatively affects exam outcomes. Beyond performance, their findings pointed out that divided attention reduces study consistency, which can create cumulative academic gaps over time.
Overall, these studies underline the superiority of deep learning models in recognizing complex behavioral signals that shape academic performance. Collectively, they emphasize that both usage intensity and the context of smartphone interactions—such as timing, emotional state, and multitasking habits—are critical in predicting and improving learning outcomes.

3 Methodology

A structured methodology consisting of five core stages:

3.1 Data Collection

The study involved gathering detailed smartphone usage logs from 5,000 university students between 2022 and 2025. The data captured not just the overall screen time but also app-specific activity, session lengths, and the frequency of screen unlocks. This broad dataset provided a comprehensive view of how students interacted with their devices on a daily basis.

3.2 Preprocessing

Before feeding the data into machine learning models, it was carefully prepared. This involved removing irrelevant or noisy records, standardizing values to a uniform scale, and converting categorical variables (like app types) into numerical formats that algorithms can process. These steps ensured that the dataset was consistent and ready for reliable analysis.

3.3 Feature Selection

From the processed dataset, meaningful attributes were extracted to represent patterns in smartphone use. Key features included average daily screen time, time allocation across different app categories (such as social media, gaming, or educational apps), the frequency of late-night phone activity, and a multitasking index that measured how often students switched between applications. These features were chosen because of their strong potential to influence academic performance.

3.4 Model Development

Three deep learning models were developed and tested:

3.5 Evaluation

To measure how well the models performed, standard evaluation metrics were applied. Accuracy provided an overall success rate, while precision, recall, and F1-score gave deeper insights into the balance between correctly identified cases and potential misclassifications. These metrics ensured a fair comparison of the models’ effectiveness in predicting student performance from smartphone usage.
image: e_a17c923fe0b2_Fig-1.png
Figure 1: Data Flow Diagram

4 Results and Discussions

The dataset table illustrates the relationship between different smartphone usage habits and academic performance (GPA) for a small group of students. A clear pattern emerges: Students with higher daily screen time, heavier use of social media, and greater night-time activity generally report lower GPAs. In contrast, students who allocate more time to educational apps and keep late-night use minimal achieve better grades. For instance, Student S003, with the highest screen time (7.8 hrs/day) and 85% night usage, has one of the lowest GPA scores (2.0). On the other hand, Student S006, who spends only 2.9 hours daily with more focus on educational apps and limited late-night use, secures the highest GPA (3.9). This highlights that balanced smartphone use, especially with emphasis on educational applications, can positively influence performance, while excessive entertainment or nighttime use undermines academic outcomes.
Table 1: Different Smartphone usage habits and academic performance (GPA)
Student_ID Avg_ScreenTime (hrs/day) SocialMedia Time (hrs) Gaming Time (hrs) EducationalApp Time (hrs) Night_Usage (%) GPA_Score
S001 6.2 3.1 1.5 0.8 72 2.3
S002 4.5 1.8 0.9 1.4 40 3.1
S003 7.8 4 2.5 0.5 85 2
S004 3.6 1 0.5 1.8 30 3.6
S005 5.2 2.5 1.2 1 60 2.8
S006 2.9 0.8 0.4 2 25 3.9
S007 6.8 3.5 1.9 0.7 78 2.2
image: e_3dbc659b50cb_Fig-2.png
Figure 2: Relationship between average daily screen time and GPA
Figure 2 shows a negative relationship between average daily screen time and GPA. As the number of hours spent on a smartphone increases, student performance generally declines. Students who use smartphones for 6–8 hours per day are clustered around lower GPA scores (2.0–2.5), while students keeping usage below 4 hours daily tend to maintain higher GPAs (3.5 – 3.9). The scatter plot provides clear evidence that prolonged daily exposure to smartphones is a risk factor for poor academic outcomes.
image: e_1860f639510b_Fig-3.png
Figure 3: The Impact of late-night smartphone activity
Figure 3 emphasizes the impact of late-night smartphone activity on academic performance. Students with night usage above 70–80% record significantly lower GPAs (2.0–2.3), while those with less than 40% night usage perform better academically (3.1–3.9). This suggests that disrupted sleep patterns and reduced rest caused by nighttime smartphone use are directly linked to underperformance in studies.
image: e_791783e1d9e2_Fig-4.png
Figure 4: Model Accuracy Comparison
The bar chart compares the predictive power of different machine learning and deep learning models. Traditional models such as Support Vector Machines (76.3%) and Random Forest (79.1%) perform moderately, but deep learning approaches significantly outperform them.
Together, the table and three figures clearly demonstrate that student digital behaviors strongly influence academic success. High engagement in social media, gaming, and night-time smartphone use lead to weaker performance, while moderate use and educational engagement promote better outcomes. Moreover, advanced deep learning models provide the most accurate predictions, offering actionable insights for educators and policymakers.
These results highlight that deep learning models not only achieve higher predictive accuracy but also provide actionable insights into how digital behavior impacts academic success. For instance, limiting late-night device activity and controlling multitasking frequency significantly improved student outcomes.
The discussion emphasizes that while smartphones are indispensable learning tools, excessive non-educational use remains a strong predictor of underperformance. Thus, institutions should develop balanced digital literacy programs integrating both self-regulation and technological monitoring.

5 CONCLUSION

This study highlights the strong connection between smartphone usage patterns and students’ academic performance. By applying advanced deep learning techniques, particularly the hybrid CNN-BiLSTM model, we were able to capture both short-term and sequential behaviors, achieving higher accuracy compared to traditional approaches. The findings suggest that while smartphones can support learning, excessive or poorly timed use—such as late-night activity, frequent multitasking, and entertainment-driven screen time—can negatively affect academic outcomes.
The results emphasize the importance of using data-driven insights to help educators, parents, and institutions create strategies that encourage healthier digital habits. As deep learning models become more powerful, the next step lies in making them interpretable so that predictions can translate into actionable guidance. Future research will therefore focus on explainable AI tools that not only predict risks but also offer personalized recommendations, supporting students in balancing technology use with academic success.

6 Acknowledgement

The authors thank, DST-FIST, Government of India for funding towards infrastructure facilities at St. Joseph’s College (Autonomous), Tiruchirappalli – 620002.

References