Emotion-Based Music Recommendation System Using

Fuzzy Logic

Submitted in partial fulfilment of the requirements for the course of

Soft Computing

Submitted by

SHUBH SHARMA (23BCE8432)

ARITRA BHATTACHARYA (23BCE8338)
HARSHIT KUMAR (23BCE8355)

Under the guidance of

Dr . POSHAM UPPAMMA

Assistant Professor Senior Grade – 1(SCOPE)

DATE : 25TH APRIL 2026

ABSTRACT Human emotions are not constant; they change based
on situations and personal feelings. Music is often used
The way computers and machines are getting smarter is to express or manage these emotions. For example,
really fast. This means that people want to have someone feeling stressed may prefer calm music, while
experiences that're just for them. This project is about a someone in a happy mood might choose energetic
music system that picks songs based on how a person's songs. Since traditional systems cannot adapt to such
feeling at that moment. It uses a camera to look at the real-time emotional changes, their recommendations
persons face and figure out if they are happy, sad, angry may not always feel relevant.
or just okay. Then it uses math to turn those feelings With advancements in AI and machine learning, it is
into types of moods like energetic, calm or sad. After now possible to recognize human emotions using facial
that it finds songs on Spotify that fit the persons mood. expressions. Deep learning models, especially CNNs,
The system has two ways of working: one is on the can detect emotions like happiness, sadness, anger, and
persons computer and uses a smart way of looking at neutrality by analysing facial features. However,
faces and the other is on the internet and is simpler so it mapping these emotions directly to music is not simple,
can be used by lots of people. The system is made with as emotions are often mixed and not clearly defined.
tools like React and FastAPI. When we tried it out To handle this, fuzzy logic is used, which allows the
people’s emotions matched the music they got better system to deal with uncertainty and partial emotions.
than with systems. This shows that using logic, which is Instead of assigning a fixed category, it interprets
a way of dealing with uncertain things and mixing it emotions in degrees and maps them to suitable moods
with psychology and web stuff can be really powerful. like relaxed or energetic. Combined with music APIs
The Emotion-Based Music Recommendation System is like Spotify, the system can provide real-time,
really good at picking music that fits a user’s state. The personalized recommendations. Overall, this approach
system uses emotion recognition and fuzzy logic to give creates a more adaptive and human-like music
users music that is just right, for them. This makes the experience compared to traditional systems.
Emotion-Based Music Recommendation System a great
way to get music that is personalized
I.1 PROBLEM STATEMENT

I. INTRODUCTION People often turn to music to reflect or change their

mood. However, choosing the right song at the right
moment can be subjective and time-consuming. There
Music plays an important role in our daily lives, as it is a need for a system that can automatically detect
strongly influences our emotions, behaviour, and how someone feels and instantly play songs that align
overall mental well-being. With the rise of digital with or improve their emotional state.
platforms and streaming services, people now have
access to huge music libraries across different genres
and moods. While this gives users more options, it also 1.2 OBJECTIVES
makes it harder to quickly choose the right music.
Traditional recommendation systems try to solve this o To develop a system that can
using methods like collaborative filtering, content- detect user emotions via
based filtering, and hybrid approaches, which mainly facial analysis.
depend on user history and preferences. However, these o To implement a fuzzy logic-based classifier that
systems usually ignore a key factor — the user’s maps emotions into music mood categories.
current emotional state. o To fetch real-time music recommendations
using public APIs based on emotional context.
 I.3 SCOPE OF THE PROJECT logic, fuzzy systems allow partial membership, making
them suitable for interpreting mixed emotional states
(e.g., 60% happy and 40% neutral). Studies have
demonstrated that fuzzy rule-based systems can
● Emotion Detection: Use of webcam input to effectively map emotional inputs to appropriate outputs
analyse facial features and detect emotions like such as mood categories.
happy, sad, angry, neutral, etc.
In the domain of music recommendation, platforms
like Spotify utilize large-scale data and machine
● Fuzzy Logic Engine: Classification of learning algorithms to suggest songs. However, these
emotional states into nuanced categories such as systems are largely dependent on listening history and
"Very Happy", "Mildly Sad", "Excited", etc., user behaviour. Recent research suggests integrating
and mapping them to music genres or moods. emotional intelligence into recommendation systems to
enhance personalization.
● Music Recommendation: Integration with Hybrid systems combining emotion detection + fuzzy
Spotify (or another music platform) to fetch logic + recommendation engines have shown
songs matching emotional state. improved user satisfaction. These systems dynamically
adapt recommendations based on the user’s current
● User Interface: A modern, responsive web mood rather than static preferences. The integration of
application to interact with the system. APIs like Spotify further enables real-time access to
vast music libraries.
Thus, the proposed system builds upon existing
II. LITERATURE REVIEW research by combining deep learning-based emotion
recognition with fuzzy logic decision-making and
real-time music recommendation, addressing the
Emotion-aware recommendation systems have gained limitations of traditional systems.
significant attention due to their ability to provide
personalized user experiences. Traditional Emotion-aware music recommendation systems have
recommendation systems primarily rely on gained attention for improving personalization by
collaborative filtering, content-based filtering, or incorporating user emotions. Yang and Chen [1]
hybrid approaches. However, these methods often fail proposed a system that utilizes contextual information,
to capture the user's real-time emotional context, including environmental factors and inferred emotional
which plays a crucial role in music preference. states, to generate music recommendations. Their
findings indicate that emotion-based approaches
Research in facial emotion recognition (FER) using provide more relevant and adaptive results compared to
deep learning techniques such as Convolutional Neural traditional keyword-based systems.
Networks (CNNs) has shown promising results.
Parashakthi and Savithri [2] developed a facial
Datasets like FER2013 and models built using
emotion recognition-based music recommendation
TensorFlow and PyTorch have achieved high accuracy
system that detects user emotions through facial
in classifying emotions such as happiness, sadness,
expressions. Their work shows that automating emotion
anger, and surprise. These systems analyse facial
detection reduces user effort and enhances the overall
landmarks and expressions to determine emotional
listening experience. Similarly, Goodfellow et al. [3]
states in real time.
applied Convolutional Neural Networks (CNNs) trained
Fuzzy logic has been widely used in systems requiring on datasets such as FER2013 for emotion classification,
handling of uncertainty and ambiguity, especially achieving improved accuracy and efficiency over
where human emotions are involved. Unlike binary traditional machine learning methods.
Recent advancements have also focused on combining III. METHODOLOGY
multiple data sources. Sharma et al. [4] integrated
facial emotion recognition with lyrics sentiment
analysis, demonstrating that the use of both visual and 3.1 Emotion Detection
textual emotional cues improves recommendation
quality. In addition, hybrid recommendation approaches ● Use a pre-trained CNN model (like FER2013 or
have been explored, where Breese et al. [5] highlighted EmotionNet) via TensorFlow or PyTorch.
the effectiveness of collaborative filtering techniques,
which, when combined with emotion recognition,
enhance system adaptability and accuracy. ● Capture user image through webcam.
Real-time systems have also been developed using
computer vision technologies. OpenCV [6] has been ● Detect facial landmarks and classify emotions
widely used for real-time facial emotion detection, (happy, sad, angry, fear, surprise, neutral,
enabling low-latency and responsive music disgust).
recommendation systems. Florence and Uma [7]
proposed a system that automatically selects music
based on facial expressions, reducing manual effort and 3.2 Fuzzy Logic System
improving user satisfaction. Similarly, Malik et al. [8]
introduced an AI-based system that maps detected Inputs: Emotion confidence
emotions directly to suitable songs, focusing on levels (e.g., 60% happy, 30%
usability and automation.
neutral,10% surprise)
Further improvements have been achieved through
hybrid filtering techniques. Gupta et al. [9] combined Rules defined for combinations of emotions (e.g., if
facial emotion recognition with both content-based and happy is high and neutral is moderate, then mood is
collaborative filtering approaches to create a more "Cheerful")
dynamic recommendation system. Kumar et al. [10] Output: Music mood (e.g., energetic, calm,
implemented a GUI-based system using frameworks nostalgic)
like Tkinter and Pygame, demonstrating effective real-
time emotion detection along with a user-friendly
interface. Rahman et al. [11] explored advanced facial 3.3 Frontend/Backend Development
expression analysis methods to enhance the precision of
emotion detection and improve recommendation ● Frontend: [Link] for dynamic UI, webcam
outcomes. integration, and result display
Finally, Zadeh [12] introduced fuzzy logic, which plays
a crucial role in handling uncertainty in human ● Backend: Fast API or Flask to handle API
emotions. By allowing reasoning with degrees of truth, calls, emotion analysis, fuzzy logic processing,
fuzzy logic enables systems to interpret mixed and Spotify communication
emotional states and map them to appropriate music
recommendations, making it highly suitable for
emotion-based systems. ● Database (optional): MongoDB or SQLite to
store logs or user preferences
IV. Architecture IV.1 Architecture Overview

Step 1: User opens web app

Step 2: Webcam captures image
Step 3: Backend detects emotion using a pre-trained
model
Step 4: Fuzzy engine analyses emotional intensity
Step 5: Backend queries Spotify with appropriate
mood filter
Step 6: Frontend displays song list or plays music

V. Results and Discussion

The developed system successfully integrates emotion

detection, fuzzy logic processing, and music
recommendation into a unified platform. The results
demonstrate the effectiveness of the proposed approach
in delivering personalized music suggestions.

🔹 Emotion Detection Performance

• The CNN-based model accurately
detected facial emotions such as happy, sad,
angry, neutral, and surprise.
• The system provided probability scores
for each emotion (e.g., 65% happy, 25%
neutral).
• Real-time detection via webcam worked
efficiently with minimal latency.

🔹 Fuzzy Logic Classification

• The fuzzy inference system effectively
handled mixed emotional inputs.
🔹 Music Recommendation Accuracy
• Songs fetched using Spotify API aligned
well with the detected emotional state.
• Compared to traditional
recommendation systems, the proposed system
showed improved user engagement.
🔹 System Performance
• The web application demonstrated
smooth interaction between frontend and
backend.
• Real-time processing was achieved with
acceptable response time.
• The system architecture proved scalable
and modular.

V.2 Graphical Analysis

V.1 Comparative Analysis
 V.3 Analysis and Justification

5.3.1 Emotion Detection Results

 Successfully detects:
o Happy

o Sad

o Neutral

o Angry

✔ Real-time processing achieved

✔ Probabilistic outputs generated

5.3.2 Fuzzy Logic Performance

 Handles mixed emotions effectively
 Generates nuanced moods:
Cheerful, Relaxed, Melancholic, Energetic
✔ Better than binary classification
✔ More human-like decisions

5.3.3 Music Recommendation Results

 Songs matched emotional state
 Audio previews enhanced interaction

5.3.4 Observations:
 Higher user satisfaction
 Context-aware recommendations

5.3.5 System Performance

Metric Result This project leverages emotion AI and fuzzy logic to
deliver an intuitive and highly personalized music
Response Time Fast (~1–3 sec) experience. By understanding the user's emotional
UI Responsiveness Smooth context in real-time, the system offers a smarter,
emotionally intelligent alternative to traditional playlist
API Integration Stable algorithms.

Scalability Good (deployed version) The system can be further enhanced by incorporating
advanced features such as voice tone-based emotion
detection to capture subtle emotional cues from speech,
5.4 User Experience Analysis along with physiological inputs like heart rate and body
language for more accurate emotional understanding.
✔ Intuitive UI Additionally, implementing multi-modal emotion
✔ Real-time feedback fusion by combining text, voice, and facial expressions
✔ Interactive audio playback can significantly improve the reliability of emotion
✔ Emotion history tracking recognition. Over time, the system can also adapt to
individual user preferences by analyzing listening
history and mood patterns, enabling more personalized
5.5 Key Findings and context-aware music recommendations.
- Emotion-based recommendation improves
personalization
VII. REFERENCES
- Fuzzy logic enhances decision-making accuracy
- Real-time AI adds dynamic adaptability
- UI/UX significantly improves user engagement
[1] I. Goodfellow, Y. Bengio, and A. Courville, "Deep
Learning," MIT Press, 2016.

5.6 Overall Findings [2] P. Ekman, "An Argument for Basic Emotions,"
• Emotion-based recommendation Cognition and Emotion, vol. 6, no. 3–4, pp.
significantly enhances personalization. 169–200, 1992.

• Fuzzy logic improves decision-making [3] A. Mollahosseini, D. Chan, and M. H. Mahoor,

under uncertainty. "Going deeper in facial expression recognition
• Real-time emotion detection adds using deep neural networks," IEEE Winter
dynamic adaptability. Conference on Applications of Computer Vision
(WACV), pp. 1–10, 2016.

[4] I. J. Goodfellow et al., "Challenges in

5.7 Limitations Observed Representation Learning: A report on three
• Accuracy depends on lighting machine learning contests," Neural Networks, vol.
conditions and camera quality. 64, pp. 59–63, 2015. (FER2013 dataset)

• Emotion misclassification can affect [5] L. A. Zadeh, "Fuzzy Sets," Information and
recommendation quality. Control, vol. 8, no. 3, pp. 338–353, 1965.
VI. Conclusion
[6] T. J. Ross, "Fuzzy Logic with Engineering
Applications," 3rd ed., Wiley, 2010.
[7] J. S. Breese, D. Heckerman, and C. Kadie,
"Empirical analysis of predictive algorithms for
collaborative filtering," UAI, pp. 43–52, 1998.

[8] B. McFee et al., "The Million Song Dataset

Challenge," Proceedings of the 21st International
Conference on World Wide Web, 2012.

[9] Spotify for Developers, "Spotify Web API,"

Available:
[Link]
api/

[10] R. W. Picard, "Affective Computing," MIT Press,

1997.

[11] S. Koelstra et al., "DEAP: A Database for Emotion

Analysis Using Physiological Signals," IEEE
Transactions on Affective Computing, vol. 3, no.
1, pp. 18–31, 2012.

[12] OpenCV, "Open Source Computer Vision

Library," Available: [Link]

[13] F. Chollet, "Deep Learning with Python,"

Manning Publications, 2017.