A Frequency Pattern Mining Model Based on Deep Neural Network for Real-Time Classification of Heart Conditions

Area of Science:

• Biomedical Engineering
• Data Science
• Artificial Intelligence

Background:

• Massive amounts of health big data are generated by IoT devices, necessitating efficient analysis for personalized medicine.
• Cardiovascular disorder risk factors require real-time identification, posing a challenge for current data processing techniques.

Purpose of the Study:

• To develop a personalized heart condition classification model using biosensor data.
• To integrate fast preprocessing techniques with deep neural networks for real-time analysis.
• To enable early recognition of cardiovascular risk situations.

Main Methods:

• Applied Fast Fourier Transform (FFT) for pulse frequency analysis and data preprocessing.
• Utilized frequency-by-frequency ratio of power spectrum for data reduction.
• Employed a deep neural network (DNN) with gradient descent for analyzing preprocessed electrocardiogram (ECG) signals.
• Trained the DNN model on pre-classified ECG signals (normal, control, noise).

Main Results:

• Achieved a data size reduction of ECG signals to 1:32 using FFT and cumulative frequency percentage.
• The deep neural network model demonstrated 83.83% accuracy in classifying real-time ECG signals.
• The proposed model significantly reduced data operation cost and processing time.

Conclusions:

• The modified deep neural network technique effectively reduces big data size and computational workload.
• This approach provides an efficient system for real-time personalized heart condition monitoring and risk assessment.
• The model aids in recognizing potential cardiovascular risks through intelligent data analysis.