Page - 312 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

3.2. Proposedsolutions forusingsensors inmentalhealthcare 3.2.1. Solutions forwearable sensors Fletcher et al. [19], have presentedwearable sensors for electro dermal activity (EDA) andmobile plethysmography, in addition tomobile phones and the supportingwireless network architecture. Wijsman et al. [21] have proposed a wearable sensor system to measure physiological signals like electrocardiogram (ECG), respiration, skin conduc- tance, and electromyography (EMG) of the trapeziusmuscles, to detect mental stress. Sevenprincipal componentswere calculated from themeasured signals, thenusedwith classifiers to detect stress or non stress conditions. Almost 80% accuracy was found. Sano and Picard [14] have represented amodel where they collected data usingwrist sensors,mobile phones and surveys, then identified features associatedwith stress, and usedmachine learning toclassify stressornostresscases.The results showedover75% accuracy. Sano at al. [25] have identified factors, fromdata collected viawearable sen- sors, that affect the academic performance of the person, then using feature selection andmachine learning they have found an association between the analyzed factors and personality types.Classificationsaccuracyusingdata collected frommobilephonesand thewearable sensors ranked from67%to92%. 3.2.2. Monitoringsystems Yamaguchietal. [15]havepresentedanindoormonitoringsystembasedoninfraredpoi- soning sensors andmagnetic sensors, inorder tomonitor thehumanactivity andbehav- ior.Butcaetal. [22]haveproposedanexperimentalmodel formonitoringenvironmental parameters connected to a cloud platform that aggregates data received from the sen- sors (e.g. body temperature, humidity from the air ). Thedata is gathered and available for furtherprocessing.Palmius et al. [23]haveproposedamulti-sensor and smartphone basedsystemthat allows remote real timemonitoringofpsychiatricpatients symptoms. 3.2.3. Mobile solutions Burns et al. [20] have developed amobile phone applicationMobilyze and a support- ingarchitecturewhich theybelieve tobe thefirst ecologicalmomentary intervention for unipolar depression, in which machine learning models predict patients’ mood, emo- tions, cognitive/motivational states, activities, environmental context, and social con- text based on phone sensor values (e.g. Recent calls, global positioning system, ambi- ent light). Promising accuracy rateswere achieved for predicting categorical contextual states (e.g.Location), from60%to91%.For states ratedon scales (e.g.Mood) thepre- dictive capabilitywaspoor. Farhan et al. [28] havepresented amulti viewbi-clustering algorithmwhich takesmultipleviewsofsmartphonesensingdataas input to identifyho- mogeneousbehavioralgroupsandthekeysensingfeatures thatcharacterize thedifferent groups.Theyhave thenemployed thekeysensing features thatdistinguish thegroups to create predictivemodels to predict the group assignment of individuals. The generaliz- abilityof themodelswasverifiedusing the supportvectormachineclassifier.Validation studies showed that the classifiers could classify individuals in the right groupwith an accuracyof87%. N.Drissi etal. /On theUseofSensors inMentalHealthcare312