In this paper, the authors described their proposed method in analyzing lifelog data in association with the environment. Tackling the problem of incomplete data, we proposed a replacement method using linear regression method which results in a normalized L2 distance score of 0.0153. Meanwhile, the authors solved the personal air quality subtask by inferring from lifeloggers' PM2.5 data, which achieves 1.0 in the arithmetic mean of absolute distance score between the predictions and the actual classes.
Along with the development of engineering and technology, more
and more personal devices such as smartphones, video cameras
and wearable sensors have come to life which provide people the
ability to easily capture every aspect of their life. On top of that,
the term lifelogging is defined to be the process of recording a
detailed trace of life passively[
The organisers generated a novel dataset called SEPHLA [
In recent years, lifelogging has gained more and more attentions
and many research works have been proposed to provide better
understanding of personal digital collections. To support, many
international benchmarking eforts have been made and various
challenges on lifelogging data were hosted, the most recent of which
is NTCIR-14 Lifelog-3 Task [
The datasets, which were utilised in these challenges, are collected by many lifeloggers who wear a passive-captured wearable camera and other tracking sensors. Each lifelogger normally generates around 1250 - 4500 images per day in association with other biometrics (e.g. heart rate, calorie), locations (GPS), physical movements and music. They share nearly the same structure with the lifelog data in the MediaEval 2019 Insight for Wellbeing Challenge. However, this challenge also considers additional information from the environment, which makes the insight more general and enables us to obtain an overview of the wellbeing among individuals. 3
From the dataset, we are provided air quality data gathered by the stations and lifeloggers’ sensors. These are extremely useful information to reconstruct missing segments of data and predict air quality index for specific areas. Besides, we also got a collection of image data recorded by the lifeloggers with corresponding visual concepts extracted from the neural network, along with the information on the checkpoints where they are asked to take pictures. However, the images which are actively taken might vary from the lifeloggers’ preferences. Therefore, it’s hard to capture and generalize the context across individuals. Based on the observation we gained, we proposed the solutions to both sub-tasks which are described in the following subsections. 3.1
In this subtask, the sequence of missing PM2.5 data is specified in each query with a starting and ending time. As the lifeloggers walked in groups, the data from others could help regenerate the missing segments. The data from the stations, however, is not quite reliable since their distance is too far from the routes and they might contribute noises to the result. Therefore, considering the NO2, O3, temperature, humidity and heartbeat data from people who share the same route with the targeted lifelogger, we build a simple linear regression model to predict the removed PM2.5 data. Specifically, let x be the 5-dimensional L2-normed feature vectors composed of five components mentioned above, and y be the targeted PM2.5 value that needs to be predicted. We construct a linear regression model y = wT x + b and apply gradient descent to find the best parameters w and b to minimize root-mean-square error, which aims to minimize the gap between model predictions and ground-truth of train data. Then the trained model is used to generate the missing PM2.5 of the targeted person in that group. As NO2, and O3 values are not almost zero for most of the times, temperature, humidity, and heart-beat are the main factors that contribute most to our predictions.
To obtain AQI for each day, we would need to first gather the air quality data. From the checkpoints of each route, we could obtain a list of GPS along the route. Then, we extracted all air quality data where lifeloggers’ GPS is closed to the checkpoints. The distance between two GPSs is calculated using the Haversine formula.
As we observed from the air quality data of each route, NO2 and O3 values are mostly zeros while PM2.5 values have some lfuctuations. Therefore, we choose PM2.5 to predict the ultimate Air Quality Index (AQI). At first, we refine the data to get the right PM2.5 data for each route by calculate the distance between the route’s GPS and collectors’ current GPS. After this step, we obtain data for 27 routes on 7 days from diferent groups of collectors. For each data on a day collected by a user, we compute its average PM2.5. Therefore, we receive many average PM2.5 values from many collectors in one day. We consider the maximum value of these average PM2.5 values as the criteria to evaluate AQI for that route on that day. Then, we average the AQI value of 7 days and re-evaluate again to infer the AQI level of the route. 4
It can be seen from the table 1 that our team (DCU) manages to achieve the 3r d highest score of approximately 0.0153 among the best submission list in the Segment Replacement sub-task. It means that our approach manages to generate relatively good prediction with low error. However, there are other solutions could provide more precise result with significantly low error.
Meanwhile, in the Personal Air Quality sub-task, our approach got the arithmetic mean absolute L1 distance score of 1.0. This means that our approach to handle the data for this task is not good and the operation that we apply to process PM2.5 data to infer AQI level is not correct. Since the data recorded from the lifeloggers walking through the route is not totally correct (as the values are almost zeros for all) and the collected data is not enough (less than 24 hours during seven non-consecutive days), we can hardly infer the right AQI level for the route.
As we do not exploit all the provided materials such as the data recorded from the stations, images and related metadata, we might miss some important features that could be used to improve our predictions. Moreover, as we rely on the users’ recorded data along the route that they pass through, the recorded values such as PM2.5, NO2, O3 are not reliable as the most of their values are zeros. These are the main factors that afect our results in both sub-tasks. In order to improve it in future work, we might need to consider additional data on the internet, which is recorded from nearby stations, to provide the missing PM2.5 values during the days to generate the correct estimation of AQI score.
This publication has emanated from research supported in party by research grants from Irish Research Council (IRC) under Grant Number GOIPG/2016/741 and Science Foundation Ireland under grant numbers SFI/12/RC/2289 and 13/RC/2106.