Browsing by Author "Rajapakse, R.A.C.P."

Now showing 1 - 4 of 4

A deep learning approach to outbreak related tweet detection
(Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Jayawardhana, B.A.S.S.B.; Rajapakse, R.A.C.P.
Social Media has become a good indicator that reflects the real-time behaviour of society. Due to the popularity of social media platforms around the world, people use to express their observations and concerns on social media. People tend to report and discuss real-world events, personal health complications, and disaster situations through these platforms. These social media data streams can be used as a means to track and detect different types of events that affect large groups of people, such as epidemics, public disorderliness and disasters. Initial outbreak reports may first appear in these platforms even before it appears in the formal sources. A mechanism to identify these outbreak-related social media posts are needed to predict the outbreak in advance. Early detection of outbreaks in advance using these social media platforms will help relevant authorities to take appropriate actions. Even though there are existing models for outbreak prediction they have limited intelligence as they have focused only on one type of an outbreak. The main objective of this research is to propose a generalized model architecture that can detect tweets related to different types of outbreaks. In this paper, we propose a deep learning model that can detect tweets that are related to different outbreaks like epidemics, public disorders, and disasters. The semantic of the tweet is very important when determining whether it might be related to an outbreak. GloVe (Global Vectors for Word Representation) word embedding are used as the feature extraction technique in this study as it can capture the semantic meanings of the tweets. Long Short-Term Memory (LSTM) which is a specialized Recurrent Neural Network (RNN) architecture that can capture long-range dependencies in sequential data like text, is used as the classification algorithm. In the process, first, outbreak-related tweets were manually collected and labelled to ensure that only true outbreak-related tweets are fed into the supervised learning model. Then the annotated Twitter dataset of 4393 tweets was curated using relevant Natural Language Processing (NLP) techniques. Pre-trained GloVe word embedding of 100 dimensions that were trained on a large corpus of tweets were then used to represent the words of the tweets. As the next step, a Deep Learning Model was trained by using LSTM technique on the curated Twitter dataset. Finally, the performance of the model was evaluated using a different dataset of 341 tweets. During this phase, the model was evaluated using performance metrics, accuracy, precision, recall, and F1-score. The proposed deep learning model performed accurately in the testing dataset with an acceptable accuracy of 89%. The results were then compared with an existing machine learning model architecture for outbreak prediction. These results indicate the effectiveness of the LSTM algorithm when detecting outbreak-related tweets and the GloVe word embedding technique when capturing the semantics of tweets. With the results of this study, we can conclude that the proposed deep learning model architecture is an accurate approach for outbreak-related tweet detection.
Detecting and Classifying Vehicles in Video Streams of Homogeneous and Heterogeneous Traffic Environments Using Gaussian Mixture Model.
(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2017) Jayathilake, M.V.M.; Jayalal, S.G.V.S.; Rajapakse, R.A.C.P.
Traffic and transportation play an important part in modern national economics. Efficient use of transportation infrastructure leads to huge economic benefits. Traffic can be classified into two main categories as homogeneous traffic and heterogeneous traffic. In transportation engineering, sufficient, reliable, and diverse traffic data is necessary for effective planning, operations, research, and professional practice. Even though, Intelligent Transport System are used to find answers for that issue still it is not yet fully successful. Many technologies have been developed to collect different types of traffic data. Traditional data collection technologies have several drawbacks. On the other hand, video based traffic analyzing has become popular. Computer vision techniques are used for detecting and classifying data in traffic videos. Those technologies are highly beneficial as it can give us more information about the parameters, easy to install and maintain and has got wide-range operation. In Computer vision, vehicle detection process has two main steps as Hypothesis Generation (HG) and Hypothesis Verification (HV). Background Subtraction is a popular method used in HG. There are several algorithms used in Background Subtraction and Gaussian Mixture Model is one of them. These methods are used in homogenous traffic situations. The objective of this study is to detect and classify vehicles from a homogenous and heterogeneous traffic video stream using Gaussian Mixture model. This study was conducted using an experimental method. Several set of road traffic videos were collected. One is collected at off peak time; i.e. 9.00am to 10.00am. At that time behavior of the traffic is similar to homogenous traffic environment. The other set of videos is collected from 7.00am to 8.30am. At that time, road traffic has no order and the traffic density is high. It is similar to heterogeneous traffic environment. After Gray Scaling and Noise reduction, the videos were submitted to algorithm based on Gaussian Mixture Model. The algorithm was implemented using Math Lab software. Vehicles are classified as large, medium and small. Manual observation results and experiment results were compared. Accurate results were observed from homogenous traffic conditions. But results in heterogeneous traffic conditions is less accurate. The Gaussian Mixture Model can be used to detect vehicles in homogenous traffic conditions successfully, but it is needed to be improved in heterogeneous traffic conditions.
A self-configuring communication protocol stack for fog-based mobile ad-hoc networks
(International Research Conference on Smart Computing and Systems Engineering - SCSE 2018, 2018) Wickramarathne, I.Y.; Jayawardhana, B.; Rajapakse, R.A.C.P.
This paper introduces a self-configuring communication protocol stack for fog-based mobile ad-hoc networks. The rapid development of Internet of Things (IoT) technologies have made mobile ad-hoc networks (MANETs) to become pervasive in our everyday lives. In MANETs, the nodes dynamically get connected and disconnected with other nodes of the network while maintaining the quality of service (QoS). However, when the devices have to contact frequently to cloud-based servers for various services and, as well as when the number of devices connected increases, the QoS could drop drastically due to high bandwidth consumption and the consequent latency. Fog computing (as well as edge computing) aims at shifting data processing and other services offered by cloud-based servers in a computer network towards the edge of the network to minimize the issues raised due to latency. Given these circumstances, combining ‘fog computing’ with MANETs seems a promising solution that enhances the QoS. However, the definition of fog computing is still debatable and, as well as the technologies are still being developed. Even though a reasonable foundation has been laid by the various concepts, there is a necessity for further research on different algorithms to meet the harsh requirements of node discovery, connectivity, communication and latency when combining fog computing with MANETs. The protocol stack presented in this paper addresses the issue of node discovery and peer-to-peer communication in MANETs in a fog network. The methodology involves a build and test approach in which the conceptual protocol stack has been implemented for messaging between mobile peers in a Wi-Fi network without connecting to the Internet.
Swarm intelligence for urban traffic simulation: Results from an Agent-based modeling and simulation study of the Sri Lankan context
(International Research Conference on Smart Computing and Systems Engineering - SCSE 2018, 2018) Amarasinghe, U.G.L.S.; Rajapakse, R.A.C.P.
Traffic congestion is a crucial issue affecting the quality of life of individuals all over the world. In a country like Sri Lanka where the traffic is mostly heterogeneous and unorganized, traffic congestion could be largely influenced by the behaviors of pedestrians and drivers. Due to the significant impact of traffic congestion to economic growth, various measures have been taken to reduce the urban traffic congestion, such as widening the roads, expanding the road network and constructing overhead bridges. However, despite all these approaches, traffic congestion still remains as a serious issue. We are of the view that the traffic congestion in Sri Lanka is largely depending on the behaviors of the pedestrians and as well as the drivers, which is something that is not adequately investigated yet. Therefore, we propose the Agent-Based Modelling and Simulation (ABMS) approach, which is a popular computational research method based on swarm intelligence to study complex social and economic systems (O'Sullivan and Haklay, 2000), for researching the impact of driver and pedestrian behavior on traffic congestion and evaluating different traffic control strategies. We used the ABMS environment called NetLogo to develop our simulator and the data collected at the Kiribathgoda junction in Western Province, Sri Lanka was to calibrate the model with accurate parameter values. Macroscopic statistics, such as the rate of traffic flow, average speeds and queue time were used to validate the model by comparing data from real traffic situations at Kiribathgoda junction with model outputs. The ultimate objective of this research is to come up with a cost-effective decision support system for administrators and policy makers to understand various reasons behind congested unorganized traffic environments in Sri Lanka and, thereby to make better-informed decisions to control urban traffic congestion.