Depression Detection Using ECG: Machine Learning

Thomala Gowthami

, Yennam Mary Poojitha

, Satri Tabita

, Yarrajodu Nandini

and Palle Sujatha

Department of CSE(AI), Ravindra College of Engineering for Women, Pasupula Village, Nandikotkur Road 518002,

Andhra Pradesh, India

Department of CSE, Ravindra College of Engineering for Women, Pasupula Village, Nandikotkur Road 518002, Andhra

Pradesh, India

Keywords: Depression, Mental Health, ECG, Machine Learning, SVM, Random Forest, CNN, LSTM, Heart Rate

Variability, Real‑Time Monitoring, Feature Extraction, Preprocessing, Physiological Signals.

Abstract: The major mental disorders affecting the society. The research utilizes electrocardiogram (ECG) signals

together with advanced Support Vector Machine, Random Forest, Convolutional Neural Network and Long

Short Term Memory based machine learning models to objectively predict depression using heart rate

variation, age, and other ECG derived features. With its promise of an early and precise detection of

depressive patterns, this method can well revolutionize mental health diagnostics with a noninvasive and cost

effective detection which can be brought into real time diagnosis and telemedicine. The major preprocessing

steps like noise filtering and normalization to improve the data quality and to automatically extract the features

as comparison to manual feature engineering are important which the key to this approach are. Additionally,

the system could be enhanced by the integration of other physiological signals, e.g. the EEG and the skin

conductance, which would allow the system to cope with ECG signal variability and to require high quality

data sets for the system to find robust and reliable implementations in real world applications.

1 INTRODUCTION

Depression is a common mental health disorder

characterized by a significant impact on one’s mood,

thought and physical well-being which can therefore

lead to major social and occupational impairment.

Traditionally, depression by us diagnose checking the

self-reported symptoms and by psychiatrists based on

a structured clinical interview. However, detecting

these early and accurately can be subjective, time

consuming and biased with the person’s experience

and they can be difficult approaches. As the need for

detecting depression, objectively, automatically and

by using data becomes even greater, for those who

may not seek professional help out of social stigma or

because of a lack of access to mental health care.

Recent biomedical signal processing advances

and machine learning methods, such detection of

depression has been enabled with physiological data

and Electrocardiogram (ECG) signals can be seen as

a possible biomarker in that process. Heart function is

closely connected to the autonomic nervous system

that, in turn, is strongly influenced by emotional

states: research indeed revealed that depression

dramatically modifies HRV and any other ECG

derived parameter. By studying these

variations in various aspects, machine learning

models are capable of classifying people as depressed

or not depressed, thus providing a reliable alternative

to such screening tests, which can otherwise be both

subjective and unreliable.

In this study, we develop a machine learning

based system for depression detection making use of

ECG signals. In the proposed work, ECG data is

collected, meaningful features are extracted, and

classification algorithms such as SVM, Random

Forest, CNN and LSTM networks are applied. The

high performance of these models is based on the use

of time domain, frequency domain and non-linear

features of ECG signals to identify depressive pattern.

In addition, deep learning techniques will be able to

extract features and features automatically reducing

the dependency on manual processing and therefore

classify better. This system can be implemented to

improve the early detection and to continuously

monitor depression in the clinical and home

environments in real time. IoT Enabled wearable

ECG devices can allow the integration in healthcare

232

Gowthami, T., Poojitha, Y. M., Tabita, S., Nandini, Y. and Sujatha, P.

Depression Detection Using ECG: Machine Learning.

DOI: 10.5220/0013925700004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

232-237

ISBN: 978-989-758-777-1

to allow remote monitoring and Telemedicine

Applications, paving the way for scalable solution of

cost effective mental healthcare. This research seeks

to fill the gap between the psychiatric assessments as

is traditionally done and the modern tech ready

mental health diagnostics through exploitation of

machine learning and physiological signal analysis to

detect depression and improve accessibility,

efficiency, and reliability in diagnosis.

2 RESEARCH METHODOLOGY

2.1 Research Area

The methodology for depression detection using ECG

signals and machine learning follows a structured

approach, beginning with data collection and pre-

processing, followed by feature extraction, model

training, evaluation, and implementation. The steps

involved in the research methodology are as follows:

Data Collection: The ECG signal data is passed from

the publicly available datasets and clinical trials of

both depressed and non-depressed people. Wearable

ECG devices or clinical grade ECG monitors record

the data (or other corresponding physiological

indicators relevant to depression) to obtain heart rate

variability (HRV).

Preprocessing of ECG Signals: However, the

observed raw ECG signals are majorly corrupted by

noise and artifacts caused by muscle movements,

respiration, and external electrical noise. Data quality

improvement is done through the application of

preprocessing techniques like bandpass filtering,

denoising and normalization in order to increase the

feature extraction.

Feature Extraction: Depression related variations

are captured from ECG signals in the form of various

time domain, frequency domain and non-linear

features. Heart Rate Variability (HRV), RR intervals,

Power Spectral Density (PSD), as well as Poincaré

plot features are key features. The machine learning

models are used with these extracted features as

input.

Machine Learning Model Training: Initial

classification is done by using Supervised learning

algorithm like Support Vector Machines (SVM),

random forest, logistic regression, etc. Moreover, use

of deep learning models (specifically Convolutional

Neural Networks (CNN), and Long Short Term

Memory (LSTM) networks) for automatic feature

extraction and higher accuracy is made. To ensure

that the model generalizes properly, the dataset are

split into training and testing sets.

Model Evaluation and Performance Metrics:

Accuracy, precision, recall, F1 score and Area Under

the Receiver Operating Characteristic Curve (AUC-

ROC) are used to evaluate the trained models. We

apply cross-validation techniques so as to reduce

overfitting and improve the generalization.

Implementation and Deployment: The model is

thus implemented in an application or embedded in

IoT enabled ECG device for real-time depression

monitoring once validated. It is possible to integrate

the system with healthcare platforms for remote

diagnosis, and also provide continuous mental health

tracking.

2.2 Research Area

ECG signals utilizing machine learning can be used to

detect depression under multiple research domains:

biomedical engineering, artificial intelligence, mental

health informatics, and wearable technology. This

research is heavily dependent on biomedical

engineering for the reason that it requires

pathophysiological signal analysis to find the

biomarkers associated with depression.

The ECG signals have the capacity to offer

insights into mental health conditions through the

heart rate and the autonomic nervous systems activity.

Biomedical signal processing techniques can be used

to extract patterns that will distinguish depressed

individuals from healthy ones, making ECG a great

tool for mental health assessment.

This research is greatly supported by the artificial

intelligence (AI) and machine learning field, as it

allows us to detect depression using AI based data

driven model. The removal of subjectivity in

traditional clinical approaches to kidney function has

been replaced by objective, accurate, scalable and

highly accurate AI driven systems. Depression is

classified using Machine learning techniques, such as

Support Vector Machines (SVM), Random Forest,

Logistic Regression using ECG derived features. In

addition, the Convolutional Neural Networks (CNN)

and Long Short Termed Memory (LSTM) models are

deep learning models that enhance the classification

performance through automatic extraction of the

complex patterns in the ECG signals and diminishing

the requirement of the manual feature extraction.

Another major research area that merges

technology and psychology as well as psychiatry to

improve diagnostic and monitoring in mental health is

mental health informatics. To reach such a high level

Depression Detection Using ECG: Machine Learning

233

of detection and tracking for depression early and

continually, digital health solutions have been

developed and used physiological signals, mobile

applications and cloud systems. This research is part

of creating a non-invasive, real time ECG based

monitoring system to incorporate in mental health

informatics that can assist healthcare professionals for

relieving depression more effectively. It improves

diagnostic precision and provides a means of

monitoring the course of depression throughout the

time course of treatment.

Moreover, wearable technology and Internet of

Things (IoT) also increase the usage of this research

by its potential for monitoring depression in real time

in life as usual. With ECG sensors integrated into

wearable devices (like smartwatches, fitness band,

etc.), they can continuously be worn without the need

of clinical visits. Typically, these are IoT enabled

devices, which transmit ECG data to cloud based

platforms that can have AI models analyzing the

signal that can in turn generate the insights related to

mental health. By developing this addiction, they can

change the way depression is detected and found to

be accessible, less stigma, and be treated on time.

3 LITERATURE SYSTEM

The authors says that the work attempts to detect

depression using ECG signals through Convolutional

Neural Networks (CNNs) and Long Short-Term

Memory (LSTM) networks, as the study of deep

learning. This presents how it uses both of those

temporal and spatial ECG data features to distinguish

between depressed and non-depressed individuals.

The results report high classification accuracy, which

serves to demonstrate that deep learning has a large

potential toward automating mental health

diagnostics. Overall, this cycle of communication

with the sensor can empower healthcare professionals

to go beyond the score put forth by the questionnaires.

 The authors discuss the statistical and

frequency domain features, the classifiers like

Support Vector Machines (SVM) and Random

Forest were utilized to identify depression

with the help of heart rate variability (HRV)

analysis. Through their study, they observe a

large correlation between lower HRV and

depression, therefore proving that HRV is an

effective marker to use when conducting

mental health assessments.

 The authors of the study introduces the use of

a hybrid feature extraction method involving

the time domain, frequency domain, and non-

linear feature of ECG signals as the robust

indicators for stress and depression. Later on,

logistic regression and neural networks are

applied for classification with this approach.

According to their results ECG is shown to be

a viable and cost effective noninvasive tool for

mental health screening with possible wider

clinical field application.

 The authors describe to improve depression

detection accuracy and since both ECG and

EEG signals contain information related to

anxiety and depression, I develop a hybrid

deep learning model that combines ECG and

EEG. This model is using CNN and BiLSTM

network to analyze both spatial and temporal

features. The research shows that combining

these physiological signals significantly

improves upon using ECG or EEG separately.

 The authors describe an IoT framework for real

time mental health monitoring with ECG

signal based sensors and cloud analytics to

assess the depression risks. To develop the

remote healthcare interventions through a

deep learning model, the model processes

ECG patterns in order to detect potential

depressive episodes.

These studies collectively highlight the

effectiveness of ECG signal analysis for depression

detection, showcasing various machine learning and

deep learning approaches for classification and

monitoring.

4 EXISTING SYSTEM

Most of the present systems for depression detection

are based on psychological assessments, self-reported

questionnaires, and clinical interview with the mental

health professional. Depression symptoms are usually

evaluated by using the commonly used standard

diagnostic tools like The Patient Health Questionnaire

(PHQ-9), Hamilton Depression Rating Scale (HAM-

D) and Beck Depression Inventory (BDI). However

these methods are subjective since they rely on the

person’s self-perception and willingness to report

honestly his or her mental state. Consequently, it can

result in misdiagnosis or underreporting, especially in

people who are reluctant to have recourse to treatment

because of the stigma or unfamiliarity they feel being

stuck at home most of the time.

A few of the systems already in place include the

use of neuroimaging techniques like functional

Magnetic Resonance Imaging (fMRI) and

Electroencephalography (EEG) in order to more

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

234

objective depression diagnosis. However, these

methods are expensive, time-consuming, and not

suitable for frequent access in the case of routine

mental health monitoring. Moreover, they are fairly

complex and require specialised equipment to be

widely used in real world scenarios.

An alternative method consists of wearable

devices and smartphone applications that monitor

behavioral patterns such as those during sleep,

exercise, and also heart rate. Artificial intelligence in

these systems is applied to spot deviations from daily

routines which could suggest depressive symptoms.

However, these methods are usually indirect and thus

limit their utility in providing precise physiological

evidence for depression and can in fact mislead to

false positives or inadequate sensitivity for clinical

diagnosis.

The existing depression detection systems do not

have required accuracy, objectivity as well as tracking

in real time. However, there is a need for a more

reliable, noninvasive, and cost effective method, thus

increasing the interest to use physiological signals

such as ECG to detect depression due to its

functionality as a direct and measurable biomarker of

the autonomic nervous system.

5 PROPOSED SYSTEM

The objective of the proposed system is to develop an

automated, non-invasive and objective, a depression

detection framework based on the Electrocardiogram

(ECG) signals and machine learning based

techniques. Such system departs from conventional

methods based on traditional self-reported

psychological assessments or expensive

neuroimaging methods and takes advantage of the

ECG based physiological markers for measuring

depression with high accuracy. The system

determines patterns of autonomic nervous system

dysfunction, a key indicator of depressive disorders

by analyzing ECG derived feature such as Heart Rate

Variability (HRV).

A wearable ECG sensor is used to obtain real time

heart signal data that the system integrates. The

preprocessing of this data is done to remove the noise

and extract the time domain, frequency domain and

the nonlinear HRV parameters. Support Vector

Machines (SVM), Random Forest and Deep Learning

models, such as Convolutional Neural Networks

(CNN) and Long Short-Term Memory (LSTM), are

applied to classify a person as depressed or Non-

depressed given their ECG patterns.

5.1 System Architecture and Results

Figure 1: System Architecture.

The system as shown in figure 1 is aimed to work

with an IoT based cloud platform to enhance the

accessibility and usability by providing the remote

monitoring and real time analysis. The machine

learning model processes and returns the instant

depression risk assessments on the cloud server using

the ECG data being securely transmitted to it. If

abnormal ECG pattern indicating depression is

detected then the system can alert the healthcare

providers or person for early intervention and

continuous mental health monitoring. The proposed

system is overall an objective, cost effective and

scalable solution for depression detection. For the

application, machine learning is employed to utilize

ECG signals, thereby replacing any level of

subjectivity in mental health diagnosis by improving

diagnostic accuracy, as well as enabling real time

monitoring whether for clinical or personal

applications.

Figure 2 shows the Total Normal & Depression

Detection Records and Figure 3 shows the Run

Existing SVM output. Figure 4 shows the Run Propose

CNN Output. Figure 5 shows the Comparison Graph

of SVM and CNN.

Depression Detection Using ECG: Machine Learning

235

Figure 2: Total Normal & Depression Detection Records.

Figure 3: Run Existing SVM Output.

Figure 4: Run Propose CNN Output.

Figure 5: Comparison Graph of Svm and Cnn.

6 CONCLUSIONS

The implementation of detection of depression from

ECG signals using machine learning offers an

innovative method to replace conventional

depression diagnosis methods by providing an

objective, non-invasive real time depression

assessment. Depression detection as it is done

traditionally, relies heavily on self-reported

questionnaires or clinical interviews both of these

aspects of depression detection can be subjective and

can result in misdiagnosis. However, ECG based

detection takes advantage of the physiological

biomarkers including heart rate variability (HRV) as

measurable indicators of mental health status. This

system is improved by machine learning models that

help to distinguish the complex patterns of ECG

signal to detect depression in earlier and more reliable

way.

The system as proposed will integrate ECG

feature extraction with advance classification

algorithms that will give high precision in

discriminating the depressed and the nondepressed

people. The system allows the analysis of these ECG

signals by utilizing techniques, for instance, Support

Vector Machines (SVM), Random Forest, and Deep

Learning models such as CNNs and LSTMs that help

improve the accuracy and efficiency of the analysis

process. In addition, the system can be used as a

wearable or IoT-based device for continuous

monitoring leading to early intervention and reducing

the amount of load on traditional healthcare facilities

and transferable to realistic applications compared to

other existing systems that typically need expensive

neuroimaging techniques or indirect behavioral

assessments. It allows remote monitoring so that

people in danger of depression are not forgotten and

get to receive health intervention when required.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

236

Furthermore, cloud based analytics brings in more

accessibility to the healthcare providers, as their

involvement in the mental health assessments is

decreased leveraging more data driven solutions.

Overall, we have made advances towards using a

machine learning driven ECG analysis for depression

detection. It then closes the gap between subjective

assessment and objective biomarkers to enable such

reliable, affordable, and preventive mental health care

solutions. Future work may consist in integrating

multimodal data sources, e.g., EEG and behavioral

data, to assist in the detection of mental health

conditions and provide an overall basis for mental

health assessment.

REFERENCES

Ansari S, Farzaneh N, Duda M et al. A review of automated

methods for detection of myocardial ischemia and

infarction using electrocardiogram and electronic

health records. IEEE Reviews in Biomedical

Engineering 2017; 10: 264-298.

Anthony H. Kashou, H. Basit, A. Malik,”ST Segment,”

StatPearls [Internet]. Treasure Island (FL): StatPearls

Publishing.

H. Wang et al.” ST Segment Change Classification Based

on Multiple Feature Extraction Using ECG,” 2018

Computing in Cardiology Conference (CinC),

Maastricht, Netherlands, 2018, pp. 1-4, doi:

10.22489/CinC.2018.253.

I. Kayikcioglu, G. Uluta§ and T. Kayikcioglu,’

Timefrequency analysis based detection of ECG ST

segment change using large feature set,” 2017 25th

Signal Processing and Communications Applications

Conference (SIU), Antalya, 2017, pp. 1-4, doi:

10.1109/SIU.2017.7960408.

I. Kayikcioglu and T. Kayikcioglu,” ECG ST change

detection using Zhao Atlas Mark time frequency distri

bution,” 2017 Medical Technologies National Congress

(TIPt E k NO), Trabzon, 2017, pp. 1-5, doi-

10.1109/TIPTEKNO.2017.8238088.

I. Kayikcioglu, C. Kose and T. Kayikcioglu,” ECG ST

segment change detection using Born Jordan

TimeFrequency Transform and Artifi-cial Neural

Networks,” 2018 26th Signal Processing and

Communications Applications Conference (SIU),

Izmir, 2018, pp. 1-4, doi: 10.1109/SIU.2018.8404266.

J. Revathi and J. Anitha, A survey on analysis of STsegment

to diagnose coronary artery disease, in Proc. Int. Conf.

Signal Process. Commun (ICSPC), Jul. 2017, pp. 211-

216.

Mohebbanaaz, Kumar, L.V.R. & Sai, Y.P. A new transfer

learning approach to detect cardiac arrhythmia from

ECG signals. SIViP 16, 1945 1953 (2022).

https://doi.org/10.1007/s11760 022- 02155-w.

Mohebbanaaz, Kumari, L.V.R. Sai, Y.P. Classification of

ECG beats using optimized decision tree and adaptive

boosted optimized decision tree. SIViP 16, 695–703

(2022). https://doi.org/10.1007/s11760021-02009-x.

Mohebbanaaz, N. G. Rani and N. P. Kumar, "AttCNNnet:

Attention Based CNN Network to Detect Seizures from

EEG subjects," 2024 IEEE 16th International

Conference on Computational Intelligence and

Communication Networks (CICN), Indore, India,

2024, pp. 800804, doi:10.1109/CICN63059.2024.1084

7370.

Mohebbanaaz, Y. P. Sai and L. V. Rajani Kumari,

"Removal of Noise from ECG Signals using Residual

Generative Adversarial Network," 2021 IEEE 8th Uttar

Pradesh Section International Conference on Electrical,

Electronics and Computer Engineering (UPCON),

Dehradun, India, 2021, pp. 1-5, doi:

10.1109/UPCON52273.2021.9667595.

Thygesen, J. S. Alpert, A. S. Jaffe, M. L. Simoons, B. R.

Chait-man, H. D. White, and the Writing Group on

behalf of the Joint ESC/ACCF/AHA/WHF Task Force

for the Universal Definition of My-ocardial Infarction,

“Third universal definition of myocardial infarction,”

Circulation, vol. 126, no. 16, pp. 2020-2035, Oct. 2012.

Y. Kan and K. Kashihara,” Automatic detection of ST

depression on ECG,” 2015 IEEE 4th Global

Conference on Consumer Electronics (GCCE), Osaka,

2015, pp. 655657, doi: 10.1109/GCCE.2015.7398704.

Y. Park, I. D. Yun and S. Kang,’ Preprocessing Method for

Performance Enhancement in CNN Based STEMI

Detection From 12-Lead ECG,” in IEEE Access, vol. 7,

pp. 99964-99977, 2019, doi: 10.1109/AC-

CESS.2019.2930770.

Depression Detection Using ECG: Machine Learning

237