<p style="text-align: left;"><span style="font-family: LMRoman12-Bold; font-size: 24pt; color: #000000;"><strong>Abstract</strong></span></p><p style="text-align: left;"><span style="font-family: LMRoman12-Bold; font-size: 24pt; color: #000000;"><br /><span style="font-family: LMRoman10-Regular; font-size: 10pt;">As technology evolves, authentication systems tend to maximize its security, while simultaneously<br /><span style="font-size: 10pt;">minimizing usability related issues. Security systems based on passwords or tokens can be easily<br /><span style="font-size: 10pt;">compromised enabling the growth of biometric systems. However, since static biometrics, such as<br /><span style="font-size: 10pt;">fingerprint or hand geometry, may be forged to circumvent systems security, different solutions<br /><span style="font-size: 10pt;">must be studied.<br /><span style="font-size: 10pt;">In this thesis the aforementioned problem will be addressed using biological signals derived<br /><span style="font-size: 10pt;">from the heartbeat, <span style="font-family: LMRoman10-Italic; font-size: 10pt;"><em>i.e. </em><span style="font-family: LMRoman10-Regular; font-size: 10pt;">Electrocardiogram (<span style="font-size: 10pt; color: #671600;">ECG<span style="font-size: 10pt; color: #000000;">) and Photoplethysmogram (<span style="font-size: 10pt; color: #671600;">PPG<span style="font-size: 10pt; color: #000000;">) waveform<br /><span style="font-size: 10pt;">signals.<br /><span style="font-size: 10pt;">Two main approaches regarding the aforementioned signals were identified, being either <span style="font-family: LMRoman10-Italic; font-size: 10pt;"><em>fiducial</em><br /><span style="font-size: 10pt;"><em>dependent </em><span style="font-family: LMRoman10-Regular; font-size: 10pt;">or <span style="font-family: LMRoman10-Italic; font-size: 10pt;"><em>fiducial independent</em><span style="font-family: LMRoman10-Regular; font-size: 10pt;">. Focusing on the latter and resourcing to a feature extraction<br /><span style="font-size: 10pt;">method exposed in the literature as AC/DCT, several data sets containing both waveforms were<br /><span style="font-size: 10pt;">generated from two public databases to evaluate the feasibility of using bio-signals in the context<br /><span style="font-size: 10pt;">of an authentication scenario in mobile devices.<br /><span style="font-size: 10pt;">Using a wristband with a <span style="font-size: 10pt; color: #671600;">PPG <span style="font-size: 10pt; color: #000000;">sensor and a mobile device, an authentication system was<br /><span style="font-size: 10pt;">simulated and evaluated. For comparison purposes, a features extraction algorithm using timedomain features extracted from Heart Rate Variability (<span style="font-size: 10pt; color: #671600;">HRV<span style="font-size: 10pt; color: #000000;">) artifacts was also implemented.<br /><span style="font-size: 10pt;">Using Random Forest (<span style="font-size: 10pt; color: #671600;">RF<span style="font-size: 10pt; color: #000000;">) as the classifier, the AC/DCT approach presented better results,<br /><span style="font-size: 10pt;">having False Positive Rate (<span style="font-size: 10pt; color: #671600;">FPR<span style="font-size: 10pt; color: #000000;">) of 7% and False Negative Rate (<span style="font-size: 10pt; color: #671600;">FNR<span style="font-size: 10pt; color: #000000;">) of 11%, and correctly<br /><span style="font-size: 10pt;">classifying 91.8% of the samples.</span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span><br style="line-height: normal; text-align: -webkit-auto; text-size-adjust: auto;" /></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></span></p>