
Open
Posted
•
Ends in 4 days
Data and Preprocessing - Dataset Source: Physionet HMC Sleep Staging dataset (EDF recordings with expert sleep stage annotations). - Subjects: Selected subset of subjects (e.g., SN001 to SN004). - Channels Used: 8 physiological channels including EEG (F4-M1, C4-M1, O2-M1, C3-M2), EMG chin, EOG (E1-M2, E2-M2), and ECG. - Sampling Rate: Resampled to 100 Hz for uniformity. - Epoch Length: 30 seconds per epoch, consistent with standard sleep scoring. - Signal Filtering: Bandpass filtering applied per channel type to remove noise and artifacts: - EEG: 0.3–45 Hz - EMG: 10–100 Hz - EOG: 0.3–15 Hz - ECG: 0.5–40 Hz - Epoch Segmentation: Continuous signals segmented into non-overlapping 30-second epochs. - Spectrogram Conversion: Each epoch converted into multi-channel spectrogram images (log-scaled, normalized, resized to 64x64 pixels) to capture time-frequency features. --- Data Balancing - Class Imbalance: Sleep stage datasets are often imbalanced, especially for stages like N1. - ADASYN Oversampling: Adaptive Synthetic Sampling (ADASYN) is applied to minority classes (N1, N3, REM, Wake) to synthetically generate samples and balance the dataset. - Target Oversampling: Each minority class is oversampled to reach 50% of the majority class size. - Effect: Improves model generalization and reduces bias toward majority classes. --- Data Augmentation - Spectrogram Augmentation: SpecAugment techniques applied during training: - Time masking - Frequency masking - Additional Augmentations: - Gaussian noise addition - Random time shifts along the time axis - Purpose: Increase data diversity and robustness to overfitting. --- Model Architecture A hybrid deep learning model combining convolutional, graph, recurrent, and transformer components: 1. CNN Encoder: - 5 convolutional layers with batch normalization and max pooling. - Extracts spatial and spectral features from multi-channel spectrograms. 2. Graph Convolutional Network (GCN): - Models spatial relationships between CNN feature map nodes. - Uses a fully connected adjacency matrix normalized for graph convolutions. - 3 GCN layers with ReLU activations. 3. Attention Layer: - Self-attention mechanism applied on GCN output feature maps. - Enhances important spatial features. 4. Bidirectional LSTM: - Processes flattened spatial features as sequences. - Captures temporal dependencies within feature maps. 5. Vision Transformer (ViT): - Projects CNN feature maps to 3 channels and upsamples to 224x224. - Applies transformer blocks with multi-head self-attention. - Extracts global contextual features. 6. Classifier Head: - Concatenates BiLSTM and ViT features. - Fully connected layers with dropout and ReLU. - Outputs logits for 5 sleep stages. --- Training Setup - Loss Function: Focal Loss to handle class imbalance by focusing on hard-to-classify samples. - Optimizer: AdamW with weight decay for regularization. - Learning Rate Scheduler: ReduceLROnPlateau to reduce learning rate on validation loss plateau. - Batch Size: 64 - Epochs: Up to 80 with early stopping patience of 18 epochs. - Device: GPU acceleration if available. --- Evaluation - Metrics: - Accuracy - Precision, Recall, F1-score (per class) - Confusion matrix - ROC-AUC for multi-class classification - Validation Split: Stratified 80-20 train-validation split. - Checkpointing: Saves best model based on validation accuracy. --- Explainability and Interpretability - Grad-CAM: Visualizes important regions in spectrograms influencing model decisions. - Integrated Gradients: Attribution method highlighting input features contributing to predictions. - LIME: Local interpretable model-agnostic explanations for sample-level understanding. - Channel Ablation: Measures importance of each physiological channel by zeroing out and observing prediction drop. - Visualization: Saves overlays of explanations and importance timelines per subject. --- Visualization and Reporting - Plots Generated: - Class distribution bar and pie charts. - Confusion matrix heatmaps. - ROC curves. - Subject-wise accuracy dot plots. - Time series hypnograms (true vs predicted). - Channel importance timelines. - Advanced plots: radar charts, treemaps, sunburst diagrams, sankey diagrams. - Purpose: Comprehensive insight into model performance, data distribution, and feature importance. --- Summary This pipeline integrates advanced signal processing, data balancing, augmentation, and a hybrid deep learning architecture combining CNN, GCN, BiLSTM, and ViT to classify sleep stages from multi-channel physiological signals. It incorporates explainability methods to provide transparency and trust in model predictions. The modular design allows easy extension and adaptation to other biosignal classification tasks.
Project ID: 39743277
19 proposals
Open for bidding
Remote project
Active 2 days ago
Set your budget and timeframe
Get paid for your work
Outline your proposal
It's free to sign up and bid on jobs
19 freelancers are bidding on average ₹1,063 INR/hour for this job

Hello, I trust you're doing well. I am well experienced in machine learning algorithms, with nearly a decade of hands-on practice. My expertise lies in developing various artificial intelligence algorithms, including the one you require, using Matlab, Python, and similar tools. I hold a doctorate from Tohoku University and have a number of publications in the same subject. My portfolio, which showcases my past work, is available for your review. Your project piqued my interest, and I would be delighted to be part of it. Let's connect to discuss in detail. Warm regards. please check my portfolio link: https://www.freelancer.com/u/sajjadtaghvaeifr
₹1,450 INR in 40 days
8.0
8.0

Hi, I’m an AI expert with professional experience in computer vision, with a proven track record of working on complex image processing and AI/ML model development. With skill sets: • Algorithm Development: Strong understanding of computer vision algorithms and techniques, including convolutional neural networks (CNNs), object detection, image segmentation and feature extraction. • Model Training & fine-tuning: Develop and train machine learning models tailored for image analysis and visual data interpretation. I have worked on some well-known models like YOLO, RCNN, U-Net, Deeplab, ViT etc. • AI Integration: Implement and integrate AI models into existing software and hardware systems, ensuring high performance and scalability. • Data Analysis: Analyze and process large datasets of images and video feeds to identify patterns, trends, and insights. • Data Handling: Experience in handling and processing large datasets, including image and video data. Familiarity with data augmentation techniques and synthetic data generation. • Performance Optimization: Optimize algorithms and models for real-time processing and ensure they can handle large-scale data efficiently. • Programming Skills: Proficient in programming languages such as Python. Experience with deep learning frameworks like TensorFlow, PyTorch, or Keras. • Tools & Libraries: Proficiency with OpenCV, scikit-image, and other relevant libraries. Experience with version control systems like Git.
₹1,000 INR in 40 days
6.1
6.1

⭐ Hello there, My availability is immediate. I read your project post on Python Developer for Deep Sleep Stage Classification Pipeline. We are experienced full-stack Python developers with skill sets in - Python, Django, Flask, FastAPI, Jupyter Notebook, Selenium, Data Visualization, ETL - React, JavaScript, jQuery, TypeScript, NextJS, React Native - NodeJS, ExpressJS - Web App Development, Data Science, Web/API Scrapping - API Development, Authentication, Authorization - SQLAlchemy, PostegresDB, MySQL, SQLite, SQLServer, Datasets - Web hosting, Docker, Azure, AWS, GPC, Digital Ocean, GoDaddy, Web Hosting - Python Libraries: NumPy, pandas, scikit-learn, tensorflow, etc. Please send a message So we can quickly discuss your project and proceed further. I am looking forward to hearing from you. Thanks
₹630 INR in 40 days
4.2
4.2

Hi, I would like to grab this opportunity and will work till you are 100% satisfied with your project. I have extensive experience in handling complex datasets like the Physionet HMC Sleep Staging dataset and implementing advanced techniques for data preprocessing, balancing, augmentation, and model architecture to ensure accurate classification. My expertise lies in creating clean, professional, and state-of-the-art deep learning models by integrating diverse components like CNN, GCN, BiLSTM, and ViT. I have successfully completed similar projects with great results. I would love to chat more about your project! Regards, Claude
₹650 INR in 7 days
3.0
3.0

I am a seasoned software developer with 13 years of experience, holding a degree from IIT Delhi. My expertise aligns perfectly with the required skills for your project. I have successfully delivered complex solutions across diverse domains with a focus on quality and scalability. I bring strong problem-solving ability, hands-on technical depth, and client-centric delivery. I am confident I can add value to your project and deliver results within timelines
₹950 INR in 40 days
0.4
0.4

Hello Manav, I am a Talented Electrical / Electronics and AI / Software Engineer with over 7 years of experience. I can work on your time zone and start working on your project immediately if you hire me. To tackle the deep sleep stage classification pipeline, I will utilize advanced signal processing techniques and deep learning models. My previous work involved developing a similar model for EEG signal classification, where I achieved impressive accuracy through dynamic data augmentation. To implement this solution effectively, I would need access to the dataset and any specific requirements you might have for the model's output formats. I would appreciate the chance to discuss the project in more detail. Best regards. ----Shohei----
₹3,076 INR in 25 days
0.0
0.0

Your project excites me, and I would love to help bring your vision to life. I understand the importance of clean, professional, and accurate classification of sleep stages from multi-channel physiological signals. With expertise in data preprocessing, deep learning, and model interpretability, I assure you of a seamless and robust solution. Although new to Freelancer.com, I bring valuable experience from numerous successful projects. Let's connect and determine if I am the right fit for your project; I am eager to assist. Regards, Justin
₹600 INR in 14 days
0.0
0.0

✨✨✨Dear Respected Customer✨✨✨ I understand you need a multi-channel sleep-stage classifier with deep learning. I will implement CNN, GCN, BiLSTM, and Vision Transformer hybrid models. Data preprocessing includes filtering, epoching, spectrograms, and ADASYN. SpecAugment, Gaussian noise, and time shifts will enhance data robustness. Focal Loss, AdamW, and ReduceLROnPlateau will optimize model training. Explainability with Grad-CAM, Integrated Gradients, and LIME will be included. Visualization dashboards with metrics, hypnograms, and channel importance plots. Thank you for hiring me.
₹1,000 INR in 40 days
0.0
0.0

Hello Manav, I can help you build a robust deep sleep stage classification pipeline based on the detailed specifications you provided. With my expertise in deep learning and data processing, I am confident in developing an efficient solution that utilizes the Physionet HMC Sleep Staging dataset to accurately classify sleep stages. I am an expert with 6 years of experience in machine learning, data augmentation, and neural networks. My past projects include building complex models that handle unbalanced datasets effectively, similar to what you described with ADASYN oversampling and SpecAugment techniques. Given the comprehensive architecture you need, I am well-equipped to implement the hybrid model combining CNN, GCN, BiLSTM, and ViT. My focus on explainability and visualization will ensure transparency in model predictions, providing insights into performance and feature importance. Thanks, Billy Bryan
₹1,757 INR in 22 days
0.0
0.0

Hi manavshah052003, I have carefully read your job description and I am confident I can handle this project successfully. With my skills and experience in Machine Learning (ML), Data Science, I will ensure the work is completed with accuracy, professionalism, and delivered on time. I always focus on providing quality results that meet client expectations and add real value. I have worked on similar projects before and understand the importance of clear communication, attention to detail, and reliability throughout the process. You can expect full dedication, regular updates, and a smooth workflow from my side. I am ready to get started right away and would be glad to discuss the project details with you. Best regards, Sidra
₹950 INR in 40 days
0.0
0.0

Hi manavshah, I've thoroughly reviewed the details of your project focusing on data preprocessing, class imbalance handling, data augmentation, model architecture, and training setup. This comprehensive approach aligns perfectly with my expertise in developing robust and interpretable deep learning models for biosignal classification tasks. This project is right up my alley, as I have a strong background in working with complex multi-channel physiological data, implementing advanced deep learning architectures, and ensuring model transparency through explainability techniques. I may be new to Freelancer.com, but I’ve delivered solid results on many similar projects off-platform. If this sounds like a good fit, I’d be happy to dive deeper into your ideas! Cheers, Leon Boshoff
₹650 INR in 7 days
0.0
0.0

Your project deserves a clean, seamless solution — and that’s exactly what I deliver. I understand the importance of data balancing and augmentation, ensuring a robust model for classifying sleep stages from multi-channel physiological signals. While I am new to Freelancer, I bring years of hands-on experience and successfully delivered similar projects off-site. I’m committed to your project's success and building a trusted profile here. I would love to chat more about your project! Regards, Romano Coetzee
₹650 INR in 10 days
0.0
0.0

I propose an advanced sleep stage classification pipeline using Physionet HMC data with 8-channel physiological signals. The approach integrates robust preprocessing, data balancing with ADASYN, and spectrogram-based augmentation to improve generalization. A hybrid deep learning model combining CNNs, Graph Convolutional Networks, BiLSTM, and Vision Transformers is trained with focal loss and adaptive learning schedules. The system is complemented by explainability methods (Grad-CAM, LIME, Integrated Gradients) and rich visual reporting (confusion matrices, ROC, hypnograms). This ensures high accuracy, interpretability, and adaptability for biosignal classification tasks.
₹1,000 INR in 40 days
0.0
0.0

Dear Sir/Madam, I have experience in writing research papers, thesis works and textbooks and course materials, matlab coding. I have experience working with complex data preprocessing, deep learning models, and explainability methods, specifically in the field of physiological signal analysis. I can help you build and fine-tune this sleep staging pipeline, ensuring the use of best practices for signal processing, model architecture, and evaluation metrics. Let’s connect in the chatbox to discuss the project further, including the budget and timeline. To know more about my experience, let's talk in a freelancer call, and I can share more details and sample works in the chatbox.
₹950 INR in 40 days
0.0
0.0

Aubervilliers, France
Member since Dec 13, 2020
₹750-1250 INR / hour
min $100000 AUD
₹600-1300 INR / hour
$1500-3000 AUD
₹100-400 INR / hour
₹600-1500 INR
$15-25 USD / hour
$15-25 USD / hour
₹12500-37500 INR
$75000-150000 CAD
$2-8 USD / hour
$1500-3000 USD
$25-50 USD / hour
$50-150 USD
$100-200 USD
$250-750 USD
₹750-1250 INR / hour
₹12500-37500 INR
$3000-5000 USD
₹1500-12500 INR
₹1500-12500 INR