Insights for Wellbeing: Predicting PM10 Values Using Stacking Ensemble Model

Abstract

In this paper, we present our ISRS-HCMUNRE team's contribution to the task Insight for Wellbeing: Cross-Data Analytics for (transboundary) Haze Prediction at MediaEval 2021. We extracted diferent types of useful attributes for the problem: the weather data, the location features, the air pollution data on the data sets provided. We applied stacking method, deep learning models, machine learning model for prediction PM10 values at diferent locations for sub task 1.

INTRODUCTION

In many countries over the world, the prediction of air pollution is getting more attention. In this study, we aim to utilize deep learning and machine learning approach using insights from data provided by the organizer to predict the PM10 value, as given in the task 1 description of the competition MedialEval 2021 [ 3 ]. This task’s primary motivation in to predict PM10 values for 3 days ahead. In this sub task, we explore the correlation between the PM10 value and the features we extracted from data set.

METHODOLOGY

Data Pre-processing

The dataset for task 1 includes weather data (temperature, humidity, wind speed. . . ) and PM10 values for three countries: Thailand, Brunei and Singapore with many data points have zero value, unreasonably large or missing. They are called anomalies or outliers, which have to be preprocessed before extracting features. We calculated average values by day for each weather attributes if they collected by hour. We dealt with missing value by filling them with mean, zero [ 1 ].

We used Box plot method to determine the outliers. By using this method, we also found the max value and the min data range and moved outlier values into this range by setting them equals the max value or the min value. The number of weather attributes data provided is diferent for each country, they are important for predicting the PM10 values.

2.2.1 Timestamp features. For Brunei dataset, the weather attributes collected by day except wind speed values collected by hour. We calculated the average wind speed by day and put all weather attributes to train models. 3

MODELING

For training an appropriate model for the problem, we split raw dataset of each country into training and testing datasets following the time value. The training data contains 80 percent of data points and the testing dataset contains 20 percent.

For predicting PM10 values, we built two models called mainmodel and sub-model. We use weather data from three days before to predict three days ahead. The main-model is a stacked model [ 2 ] including three deep learning models: Long Short-Term Memory (LSTM)[ 4 ], Bi-directional (Bi-LSTM), Gate Recurrent Unit (GRU)[ 5 ] and a machine learning model: Linear Regression. The sub-model is a LSTM model that we used to predict PM10 values if there are missing weather data points in result data file. The data for training sub-model is only PM10 values from days before. 4 To enhance the prediction result, we employed the stacked generalization technique. The stacked model has two levels: level 0 and level 1. The level 0 data is the training dataset inputs and level 0 models learn to make predictions from this data. The level 1 input data is the output of level 0 models and the single level 1 model, or meta-learner to make predictions form this data.

We utilized three models: LSTM, Bi-LSTM, GRU, as level 0 models. We used Linear Regression model as level 1 model for final prediction.

5 PERFORMANCE METRICS

For evaluating the performance on the proposed methods, we use root mean square error (RMSE), mean absolute error (MAE), as follows: v u t 1 Õ

=1 =

( − ˆ)2 1 Õ = ( ) =1 | − ˆ | Where ˆ is the ℎ predicted value from model, ¯ is the average of observed values, and is ℎ observed value, (i = 1, . . . , N).

6 RESULTS AND DISCUSSION

After extracting the necessary information, we evaluated four deep learning models: LSTM, Bi-LSTM, GRU, single variable LSTM and stacking model on testing data set. The table 1,2,3 show the RMSE, MAE score for each air quality station.

There are a lot of data points missing in Thailand and Singapore data sets leads to the testing result in poor performance. Lifelogging for wellbeing

References

[1] Denis Cousineau and Sylvain Chartier . 2010 . Outlier detection and treatment: a review . International Journal of Psychological Research ( 2010 ), 58 - 67 .

[2] Dat Q. Duong , Quang M. Le , Tan-Loc Nguyen-Tai, Hien D. Nguyen , Minh-Son Dao , and Binh T. Nguyen . 2021 . An Efective AQI Estimation Using Sensor Data and Stacking Mechanism . In New Trends in Intelligent Software Methodologies, Tools and Techniques - Proceedings of the 20th International Conference on New Trends in Intelligent Software Methodologies, Tools and Techniques , SoMeT 202 , Cancun , Mexico, 21 - 23 September, 2021 (Frontiers in Artificial Intelligence and Applications) , Hamido Fujita and Héctor Pérez-Meana (Eds.) , Vol. 337 . IOS Press, 405 - 418 . https://doi.org/10.3233/FAIA210040

[3] Asem Kasem , Minh-Son Dao , Efa Nabilla Aziz, Duc-Tien DangNguyen , Cathal Gurrin , Minh-Triet Tran , Thanh-Binh Nguyen , and Wida Suhaili . Overview of MediaEval 2021 : Insights for Wellbeing Task Cross-Data Analytics for Transboundary Haze Prediction .

[4] Jung-Hwan Park , Seong-Joon Yoo , Kyung-Joong Kim , Yeong-Hyeon Gu , Keon-Hoon Lee , and U-Hyon Son . 2017 . PM10 density forecast model using long short term memory . In 2017 Ninth International Conference on Ubiquitous and Future Networks (ICUFN) . 576 - 581 . https: //doi.org/10.1109/ICUFN. 2017 .7993855

[5] Guang Yang , HwaMin Lee , and Giyeol Lee . 2020 . A Hybrid Deep Learning Model to Forecast Particulate Matter Concentration Levels in Seoul, South Korea . Atmosphere 11 , 4 ( 2020 ). https://doi.org/10. 3390/atmos11040348