Deep Emotion Recognition using Wearable Sensors
Emotion recognition is an emerging interdisciplinary field that integrates methodologies from affective
computing, sentiment analysis, signal processing, and machine learning. This project focuses on classifying
emotions such as amusement and stress using physiological signals like heart rate and skin conductivity,
collected via wearables. To build the model, a hybrid approach was employed that combines various deep
architectures, including Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. This
approach is capable of accurately capturing changes in physiological signals to identify specific emotional
The performance of the model has been rigorously evaluated using a variety of metrics, including accuracy,
precision, recall, and F1-score. To fine-tune the model's performance, hyperparameter tuning was conducted
grid search and Bayesian optimization techniques. The model achieved an accuracy rate of 92% under a 5-fold
cross-validation setting. As wearable technology continues to evolve, this project serves as a significant
contribution to the development of real-time affective computing systems that could be integrated into future
generations of a wide range of wearable products, from smartwatches to health monitors.
E4 TimeStamper : A GUI application for automatic timestamping and analysis of physiological signals collected
from Empatica E4 wristbands
E4 TimeStamper is a user-friendly GUI application designed to
facilitate researchers in adding timestamps to
physiological signal data obtained from Empatica E4 wristbands. This
tool allows for seamless extraction and
precise timestamping of files, all in accordance with the chosen timezone and preferred date & time format.
E4 TimeStamper, adopted and implemented by researchers worldwide in
their studies,1 is available for both Windows and Mac
For more information on how to download and use the app, please go to the GitHub page.
View on GitHub
Python Implementation of OpenIntro Statistics Labs
We developed the Python labs for OpenIntro
Statistics, an open-source textbook for introductory statistics used at many universities (from
Colleges to the Ivy League)
the world, to promote the understanding and application of statistical data analysis using Python.
The labs are officially available on the OpenIntro Statistics website.
BBL-PhysioDB : Data Hub for Entrepreneurship Assessment and Development Workshop
BBL-PhysioDB is a robust database, developed with Django and PostgreSQL, for the Entrepreneurship Assessment
and Development Workshop studies conducted at RMIT's
Behavioural Business Lab (BBL). The database can effectively
manage data collected from over 150 participants, including entrepreneurs, artists, and professionals with
diverse backgrounds. It has the capability to store extensive data collected through multiple lab experiment
sessions, including physiological signal data recorded through E4 wristbands from Empatica, along with 200+
questionnaire responses providing demographic information and insights into participants' entrepreneurial
The database offers a user-friendly interface that simplifies navigation and enables robust search
Researchers can seamlessly query the database to explore findings from BBL studies. The database scheme can be
leveraged by other researchers to address their own research needs.
Melbourne Property Sales: A Visual Exploration of Housing Market Dynamics
Melbourne Property Sales is an interactive dashboard with a detailed visual overview of the city's real
trends, based on historical sales data. The dashboard offers an range of visualizations, including average
property prices over time, distributions of key property features, and regional price comparisons for
various housing types.
These visualizations highlight patterns on the Melbourne housing market dynamics, such as seasonal pricing
trends and the common characteristics of sold properties. For further analysis and insights, please refer to
the links provided.
Advancing Water Safety : An Early Warning System to Monitor and Evaluate Drinking Water Quality
An early warning system was developed that can enhance the safety and quality of drinking water in Turkiye
identifying hazardous contaminants such as E. coli O157:H7, anthrax, and ricin. Utilizing a hybrid approach
that integrates various statistical methods like Z-score analysis, moving averages, control charts, and
voting, the system was designed to provide real-time monitoring and detect unexpected levels of multiple
parameters, including temperature, pH, total organic carbon, conductivity, oxidation-reduction potential,
chlorine, and dissolved oxygen.
The idea with employing a multi-tiered approach is to facilitate early intervention for minor deviations
while enabling immediate action for more severe anomalies. It has the potential for wide-ranging
not just limited
to municipal water treatment plants but also extending to main storage and distribution lines, thereby
as a critical tool with the capability to safeguard public health and environmental integrity.
DurCalc : Hassle-Free Calculation of Date and Time Durations
is a web app designed to
effortlessly calculate the duration between dates and/or times.
With its user-friendly interface and intuitive functionality, DurCalc streamlines the process and calculates
durations without any fuss.