This project addresses the challenge of monitoring large, dynamic classrooms by proposing a privacyoriented multimodal data acquisition system tailored for MMLA. Traditional learning analytics rely on unimodal data and fail to capture complex classroom interactions. In contrast, MMLA leverages multiple data sources to better understand learning behaviors. Current systems lack adaptability, userfriendliness, and privacy considerations, impeding their integration into classrooms. The proposed system comprises static and dynamic nodes, with dynamic nodes worn by individuals and static nodes strategically placed in classrooms. Data features, selected on MMLA relevance, are transmitted wirelessly to the static node for storage and analysis. Privacy is prioritized by avoiding sensitive data collection and adhering to GDPR guidelines. The design ensures adaptability, supporting additional sensors and seamless integration into various educational settings. This foundational system enables future research while addressing ethical and technical challenges in large-scale classrooms.

This thesis, conducted as part of the Bachelor Graduation Project at TU Delft, focusses on designing the communication and storage components of a data acquisition system for Multimodal Learning Analytics (MMLA). The goal is to create an adaptable system that improves learning outcomes in large, dynamic classrooms such as Tellegen Hall. Current systems often fail to address issues of privacy and integration, limiting their effectiveness in real-world applications. Theproposeddesignemployswearabledynamicnodesfordatacollection, andutilisesstaticnodesand arootnodeforwirelesscommunication. WirelesscommunicationusestheESP-NOWwirelessprotocol, and data is stored in a time-series database, InfluxDB, running on a Raspberry Pi 5. A Graphical User Interface(GUI) built in Grafana, allows monitoring and visualises the data. The system was tested for reliability and performance, showing a high success rate of 99.93% in data transmission. It enables a network consisting of, theoretically, up to 200 nodes. Future improvements include real-time analytics and data security mechanisms.

Atrial fibrillation is a common cardiovascular disease, affecting the regular beating of the heart through chaotic contraction of the heart's upper chambers. On its own, the condition—increasingly prevalent among the elderly—is not life threatening, but it leads to an increased risk of stroke and heart failure. As of yet, there is no consensus on the physiological mechanisms responsible for initiating and sustaining atrial fibrillation. A more detailed view of cardiac activity would improve understanding of the disease, making earlier diagnosis possible and improving options for treatment.

The contraction of the cardiac muscles is governed by electrical signals propagating through the tissue. Cardiac activity can be monitored with a high spatial resolution by measuring the electrical potential directly on the epicardium of the heart during open-heart surgery, using an array of closely spaced electrodes. From these electrograms, estimating the time of local activation of the cardiac tissue underneath each electrode provides a quantitative way of mapping the mechanisms of atrial fibrillation. Various methods exist to estimate the activation times, but the complex signals that are typical of atrial fibrillation make it difficult to obtain accurate results. This thesis proposes combining two existing methods for estimating the local activation times. Based on a model of the electrogram as a spatial convolution of local transmembrane currents, an inverse problem is formulated and solved, resulting in a less opaque view of the cardiac activity at the electrode locations by attenuating distant disturbances and emphasizing local activity. The deconvolution output is fed to the second step, where cross-correlating certain pairs of signals gives an estimate for the mutual time delay in local activation. A graph representation of the electrode array is used to define neighbor order and decide which signal pairs are correlated. The set of pairwise time delays this produces is then converted to an estimate for the local activation times, using a least-squares estimator.

The proposed method is evaluated using different simulated cardiac settings. In a setting with one stimulation source, earlier results of the deconvolution and cross-correlation methods are confirmed, and the proposed method is seen to produce a slightly lower mean error than reference methods. In the higher-complexity triple-source setting, the latter effect is again visible. Reinforced by the performance of the different methods in increasingly noisy settings, the main merits of the proposed method for the estimation of local activation times can be said to be found in the form of increased consistency, not significantly improving on the accuracy of existing methods.

In this work, we propose an IoT smart classroom framework which will be the base for smart classroom and hybrid teaching applications. The prototype introduces sensor readout capabilities, which can be used to monitor: temperature, humidity, and loudness in a classroom. Furthermore, the prototype introduces capabilities to collect and display measurements. Additionally, this prototype will assist an educator in a hybrid teaching setup by offering additional functionalities. With this prototype, the educator is able to control whether the students online are able to hear physically present students or not. The prototype also introduces a way to notify the educator of any questions from students attending online. The goal of this thesis is to make the first steps to improve the hybrid teaching environment by reducing the workload on the educator and by making it easier for online students to interact with the educator.

The project is divided into three submodules. Each submodule designs a different part of the total project. This thesis focuses on the hardware of the system. The hardware that is used for audio recording, audio controlling, question indicating and climate monitoring. The other two subgroups discuss IoT communication, data storing and graphical displaying of data.

In this work, we propose an IoT smart classroom framework which will be the base for smart classroom and hybrid teaching applications. The prototype introduces sensor readout capabilities, with which temperature, humidity, and loudness in a classroom can be monitored. Furthermore, the prototype introduces capabilities to collect and display measurements. Additionally, this prototype will assist an educator in a hybrid teaching setup by offering additional functionality, namely notifying the educator of online questions and giving an overview of environmental measurements in the classroom. With this prototype, the educator is able to control whether the students online are able to hear physically present students and vice versa. The prototype also introduces a way to notify the educator of any questions from students attending online. The goal of this prototype is to make the first steps to improve the hybrid teaching environment by reducing the workload on the educator and by making it easier for online students to interact with the educator. This prototype is split into three theses, focusing on sensor hardware, the Bluetooth mesh network, and the server respectively. This work will have a higher focus on designing and creating a server for this prototype, which will collect and display sensor measurements.

This report details the design and implementation of a subsystem providing a web-based platform for smartwatch data acquisition. A smartwatch application is developed to read out the accelerometer, gyroscope and heart rate sensor on the smartwatch and transmit the sensor data to the platform. A platform is developed to receive the data and store it in a database. Recordings can be downloaded from the platform to use in research of human activity. The platform presents the type of activity the smartwatch user is doing using machine learning models, by integrating the other subsystem. Multiple tests have been performed to analyse and improve the performance of the system. The battery life of the smartwatch has been tested using various settings in the smartwatch application to determine the most power-efficient settings.

This report entails one of two subsystems in a joint project to provide a web-based platform for smartwatch data acquisition, for applications in healthcare. In this work, we design and implement algorithms for human activity recognition using various machine learning approaches and test them on data acquired online as well as using our own developed platform. Together with the web-based platform, this provides a solid base for more research using data gathered from smartwatches. The human activity recognition is implemented first using a classical machine learning approach
with feature extraction and a random forest classifier. Next, both convolutional neural network and a recurrent neural network are implemented using Tensorflow [1]. We further perform several tests to investigate: (i) the optimal segment size with respect to classification accuracy, (ii) the effect of filtering and preprocessing on the classification results, and (iii) the best classifier for activity detection.

When symptoms of atrial fibrillation (AF), a common cardiac arrhythmia, are experienced, a Holter monitor or event recorder is used for official diagnosis. Apart from the fact that these devices are experienced as inconvenient, AF can already manifest damage in a pre-symptomatic phase. This thesis is aimed at developing a method for recording heart activity using a wearable device to permit convenient early detection of AF. For this, heart activity is measured continuously by means of photoplethysmography (PPG). A classification algorithm is used to detect AF episodes in the PPG recording. If the algorithm suspects AF, a limb lead I ECG recording is requested from the user. The ECG recording can be analyzed by a clinician for official diagnosis. The Maxim Integrated Max86150 chip is used for the implementation of PPG and ECG. Acceleration data is gathered by means of the Adafruit MMA8451 accelerometer to allow for detection of motion artefacts. These sensors and the data they retrieve are controlled and processed by the ARM Cortex-M7 microcontroller. From the results, PPG recordings have a higher quality when infrared light is used as compared to when red light is used. However, both types of recordings are of sufficient quality for monitoring the heart rate accurately when in stasis. Although complete functionality of the system could not be verified, the results are promising for future work.

Atrial fibrillation (AF) is the most common type of cardiac arrhythmia occurring in around 0.5% of the world population. AF is characterized by the rapid and irregular beating of the atrial chambers of the heart, which can cause lead to strokes and other heart-failures. To prevent these consequences the early detection of AF is paramount. Using photoplethysmography (PPG) heart activity can be measured from which the inter-beat-interval (IBI), the time between heart beats, can be estimated. Using data collected by a PPG sensor the aim is to classify the heart activity as either AF or Normal Sinus Rhythm in real time using machine learning and collect the outcomes for further analysis by medical professionals. For this a classification method is suggested which is able to be implemented on an ARM based processor. Using a Support Vector Machine and 10 features derived from the IBI's and the PPG signal this algorithm achieves the following accuracy metrics: balanced accuracy = 0.853, sensitivity = 0.850, specificity = 0.856 and Matthews Correlation Coefficient (MCC) = 0.643. Compared to similar studies these results are substandard and should be improved.

The performance of wearable Photoplethysmogram (PPG) sensors is highly influenced by noise. This thesis describes the methods and results of designing a filtering systemfor PPG in context of Atrial Fibrillation (AF) detection. The developed work is an adaptive filtering system combined with a robust heart rate detection mechanism for validation of the proposed method. Additionally, the heart rate estimation can potentially be used to detect AF episodes using machine learning. Research has been done regarding an optimal reference signal for the adaptive filtering structure. Accelerometer data, being commonly used as reference signal for the noise did not showgood correlationwith the motion induced artefacts in the signal. Therefore, a reference for the signal component is generated from the PPG itself, which is achieved by applying a narrow bandpass filter. Here the center frequency is determined from an autocorrelation of the signal in a sliding-window. The optimal settings for the sliding window in AF context were found to be 2 seconds with 80% overlap. Furthermore, a comparison is made between NLMS, RLS and Kalman adaptive algorithms, in which RLS showed the best overall performance. The validation of the filtering structure is based on peak detection from the enhanced signal compared with the ECG reference peaks. The results indicates that the system significantly improves the heart rate error in signal disturbed by noise and during AF episodes.