Adaptive user interface for mobile devices based on activity recognition

Companies giving service to their customers using mobile applications have been vastly increasing in the last five years. Mobile applications targeting both end users and corporates have become more important in main business processes of companies in nearly all sectors. Financial transaction counts triggered from mobile devices have rapidly increased and have finally overrun transaction counts that were performed using traditional financial channels such as ATMs, bank branches and physical POS devices. In order to aim various customer profiles, companies tend to develop as many feature as possible targeting the limited space that mobile devices work with. Furthermore, due to the profit maximization strategy of companies, these features and options are prioritized using their profit ratios and amounts. New features, campaigns and advertisements also have high priority than existing and widely known ones. Since mobile device and mobile application capabilities have increased in the past years, the way that users behave in daily activities has also changed. Mobile applications are being used for not only entertainment and information collection, but also for daily operations such as checking emails and paying bills. Moreover, people often use mobile applications while performing other activities. We can categorize end users' mobile application usage needs in two categories. First of all, when users are in hurry or when they are performing other physical activities, they need to complete their tasks quickly, easily and without any problems. Secondly, in their free time or when they are not in a hurry, they want to have more options, be informed about new features and opportunities and experience new and innovative user interfaces. We have figured that a mobile application should meet a user's changing needs in different contexts without negatively affecting a company's business and profit targets. In order to achieve this, an adaptive user interface may be of service. We have planned such an adaptive mobile application that identifies the user's current situation using the activities she has been performing. After obtaining user activity data with the help of the sensors on the mobile device, machine learning algorithms are to be used for determining and classifying user activities. Human activity recognition solutions are mostly implemented on security, health, child care, elderly care and sports areas. Walking, running and personal care activities are frequently classified. There are numerous works that compare device types, sensors used and the body part that the user carries the device on. Also, some studies compare different machine learning methods. Activity recognition solutions have several application areas such as adaptive user interface personalization, user dependant content filtering and targeting, user training and instruction in the initial usage of the application, and user behaviour prediction that aims to prevent the user from performing some tasks. Within our thesis study, user activity data was collected from end users with the help of a native Android mobile application. Every user installed the mobile application on her phone personally. After the collected data was uploaded to the server, feature extraction was applied and 45 features were extracted. These features consist of maximum, minimum, mean, standard deviation and mean absolute deviation values of device sensor data. Sliding Windows method was used for feature extraction. Afterwards, Best First search method was used and Correlation-based Feature Selection method was applied in order to determine the features to be used. 25 features were found suitable and usage ready. Random Tree, Decision Table and Bayesian Network models were trained using the data, and classification models were obtained. After performing cross validation on all of the three algorithms, models were validated using an independent test data set. Test results were evaluated and the best performing model was identified. Among three different methods (Random Tree, Decision Table and Bayesian Network) Bayesian Network was found the one with the more success on detecting the right activity. In this study, adaptive user interface usage aims to enhance user experience for users that use the mobile application while performing other activities. In order to obtain end user expectations, a set of questions were asked to the testers that were selected for understanding user needs. According to the answers of the questions, adaptive user interface design requirements were settled. Using these requirements, a native Android application that has both static user interfaces and adaptive dynamic user interfaces was implemented. The adaptive user interfaces are designed to be used during physical activity. Both versions of the application were provided to the testers. The users were asked to perform a set of banking operations using the mobile application. The tests were performed during five different activities, including standing, walking, climbing up the stairs, climbing down the stairs and running. All user operations were recorded as video files. Afterwards, a user satisfaction survey was applied to the testers. Video recordings of the operations performed using the two versions of the mobile application were examined, and durations were compared. Survey results were analysed. In order to evaluate the quantitative results of the tests, durations of the tasks performed by testers are calculated using the video files recorded during the tests. The durations for each activity in both static user interface and adaptive dynamic user interfaces were analysed. The results showed that the use of adaptive dynamic user interface during the activity helped users to achieve their task in less time comparing to use of static user interface. Another analysis is made on the results of the user satisfaction survey. These results showed that majority of the testers are highly satisfied by the experience and stated that they would prefer to use mobile applications with the design provided by the adaptive user interface. This work is applicable to all kind of mobile application and mobile device. The use of adaptive user interfaces would be helpful in music playing applications which are used during exercises, navigation applications which are used while walking or driving car or all social media application that users like to check during their other activities. On the other hand, the financial applications which contains processes taking long time, which require high attention or which contain urgent tasks are also an application area of adaptive user interfaces.