Recent increase in widespread use of facial recognition technologies have accelerated the utilization of demographic information, as extracted from facial features, yet it is accompanied by ethical concerns. It is therefore crucial, for ethical reasons, to ensure that algorithms like face recognition algorithms employed in legal proceedings are equitable and thoroughly documented across diverse populations. Accurate classification of demographic traits are therefore essential for enabling a comprehensive understanding of other algorithms. This thesis explores how classical machine learning algorithms compare to deep-learning models in predicting sex, age and skin color, concluding that the more compute-heavy deep-learning models, where the best performing models achieved an MCC of 0.99, 0.48 and 0.85 for sex, age and skin color respectively, significantly outperform their classical machine learning counterparts which achieved an MCC of 0.57, 0.22 and 0.54 at best. Once establishing that the deep-learning models are superior, further methods such as semi-supervised learning, a multi-characteristic classifier, sex-specific age classifiers and using tightly cropped facial images instead of upper-body images were employed to try and improve the deep-learning results. Throughout all deep-learning experiments the state of the art vision transformer and convolutional neural network were compared. Whilst the different architectures performed remarkably alike, a slight edge was seen for the convolutional neural network. The results further show that using cropped facial images generally improve the model performance and that more specialized models achieve modest improvements as compared to their less specialized counterparts. Semi-supervised learning showed potential in slightly improving the models further. The predictive performances achieved in this thesis indicate that the deep-learning models can reliably predict demographic features close to, or surpassing, a human.

The ownership and demand for fitness trackers, smartwatches, and wrist-worn deviceshave been increasing globally. These devices offer various features such as measuringphysical activity, sleep monitoring, and health-related measurements like heart rate andheart rate variability using PhotoPlethysmoGraphy (PPG). However, research indicatesthat PPG measurements are less accurate on darker skin compared to lighter skin due to thehigher presence of melanin, a light-absorbing substance in dark skin.This thesis addresses the impact of melanin on the accuracy of heart rate measurements ondifferent skin colors using four commercial smartwatches, Apple Watch Series 5, FitbitCharge 2, Xiaomi Miband 3, and Sony mSafety. The study involves analyzing the accuracyof these smartwatches on individuals with varying skin colors while controlling forexternal factors. The collected data from the smartwatches are compared to a referencesensor that uses electrocardiography (ECG) measurements with electrodes placed aroundthe chest. Three different tests are conducted wearing the devices, with no movement,while walking, and with circular hand motions.The tests were conducted on twelve participants representing the 6 different skin typescategorized using the Fitzpatrick scale. With the presented results in this thesis, it wasconcluded that the 4 smartwatches' measurement accuracy does not seem to be dependenton specific skin types. Ranked in terms of Mean Absolute Error (MAE), the Apple Watchshowed the lowest value, followed by Xiaomi, Fitbit and the mSafety and that the accuracydid not depend on skin color. / Ägandet och efterfrågan av smartklockor har ökat globalt. Dessa enheter erbjuder olikafunktioner som mäter fysisk aktivitet, övervakning av sömn och hälsorelaterade mätningarsom puls och hjärtfrekvensvariabilitet med hjälp av PhotoPlethysmoGraphy (PPG).Forskning har dock visat att PPG-mätningar är mindre noggranna på mörkare hud jämförtmed ljusare hud på grund av den högre närvaron av substansen melanin, ettljusabsorberande ämne.Syftet med detta arbete är att undersöka påverkan av melanin på noggrannheten i pulsmätningar på olika hudfärger med hjälp av fyra kommersiella smartklockor: Apple WatchSeries 5, Fitbit Charge 2, Xiaomi Miband 3 och mSafety av Sony. Arbetet inkluderar enanalys av dessa smartklockors noggrannhet på personer med de olika hudfärger samtidigtsom externa faktorer som kan påverka noggrannheten kontrolleras. De insamlade data frånsmartklockorna jämförs med en referenssensor, Polar band, som använderelektrokardiografimätningar (ECG) med elektroder. Tre olika tester utförs med enheterna,utan rörelse, medan man går och med cirkulära handrörelser.Testerna utfördes på tolv deltagare som representerar olika hudtyper kategoriserade enligtFitzpatrick-skalan. Utifrån de presenterade resultaten i denna avhandling drogs slutsatsenatt de fyra smartklockornas mätnoggrannhet inte verkar vara beroende av specifikahudtyper. Rankade i termer av Mean Absolute Error (MAE) visade Apple Watch det lägstavärdet, följt av Xiaomi, Fitbit och mSafety.