Paper Info
Title
Two-Stages Occupancy Number Detection Based on Indoor Environment Attributes By Utilizing Machine Learning Algorithm
Abstract
As the building energy conservation awareness increased, many sensors and methods utilized as an effort to achieve energy saving. There are many sensors already utilized. However, each sensor has its disadvantages. Therefore, the indoor environment sensor seems to be suitable because it is a non-intrusive method, non-terminal based method, and the variables have a high correlation with the presence of the occupant. Time-series prediction of the indoor environment attributes conducted using LSTM. LSTM has proven to give high-accuracy prediction result for the three different testing datasets the accuracy as follows: 89.3%, 95.1%, 93.6%. Hence, after prediction conducted to obtain occupant number from the indoor environment features the comparison of six machine learning classification methods which are: SVM, RF, DTC, MLP, k-NN, ANN also conducted. Three datasets used to test the model, both original data and oversampling data used. Thus, to address the real-world problem condition where imbalanced class occurred due to uncertainty. Overall, SMOTE oversampling boost all of the machine learning performance while obtaining occupant number. SMOTE Algorithm contributes significantly to all of the machine learning while predicting multiclass imbalanced dataset (Dataset-2 and Dataset-3). For the binary problem (Dataset-1), SMOTE's contribution to the machine learning model performance is not significant. Three of the machine learning algorithms which are RF, DTC, and k-NN produce good accuracy consistently. Thus, those three algorithms utilized to obtain occupant numbers after Indoor Environment attributes which predicted using LSTM. From the three algorithms utilized RF, and KNN outperforms DTC in terms of accuracy consistency for three oversampled datasets. For three dataset RF with SMOTE obtain accuracy as follows: 97.8%, 63.2%, and 79.1%. On the other hand, KNN with SMOTE produces accuracy as follows: 96.8%, 61.4%, and 81.5%.
Year
DOI
Venue
2019
10.1109/iFUZZY46984.2019.9066241
2019 International Conference on Fuzzy Theory and Its Applications (iFUZZY)
Keywords
DocType
ISSN
indoor environment,occupancy number detection,energy conservation,data mining,classification,prediction,machine learning
Conference
2377-5823
ISBN
Citations 
PageRank 
978-1-7281-0841-4
0
0.34
References 
Authors
3
6
Name
Order
Citations
PageRank
Jesica Elizabeth Marchelina100.34
Shuo-Yan Chou253748.50
Vincent F. Yu342427.32
Anindhita Dewabharata421.42
Vicky Chrystian Sugiarto500.34
Irene Karijadi600.68