• • Lakshitha Deshapriya

  • Ishan Madhusanka

  • Ishani Paranawithana

  • Titus Nandakumara

The complex process of training placement process for undergraduates was handled by manually over the past years. Collecting students CVS and preferences for companies, collecting company requirements, grouping students for companies and scheduling interviews were the main activities of

the process. Handling the process manually consumes a considerable amount of time and result is also not much accurate. Therefore, as the solution, an intelligent training portal with the characteristics of recommendation engine was introduced to automate the process. The system consists two significant components; the front-end implementation and the back-end implementation. The back-end implementation includes both the

authentication server and resource server where important decisions are made through machine learning. Various classification algorithms and data that was collected throughout past years were used to train the system to classify the students for companies which have a higher probability of accepting them.

• • Pranavan Somaskandhan

  • Gihan Wijesinghe

  • Leshan Bashitha Wijegunawardana

IPL is a franchise system based, annual cricket tournament. IPL deals with millions of dollars. This imposes high pressure on team owners to search victories, which depends on team performance. The aim of this research is to assemble an optimal cricket team within a given budget to enhance the winnability. Several efforts have already been taken to address this problem without much success. They focused on identifying different performance metrics based on their domain knowledge of cricket. Essentially, it is critical to find the right set of metrics that would lead to assemble a team with

the highest chance of winning. The proposed solution is, rely on statistical analysis and machine learning while minimizing the use of domain knowledge. This study has started with gathering and refining necessary data. Then an optimal set of attributes has been identified, which impose the high impact on the end results of a cricket match. For this, Classification algorithms have been used and SVM gave the best accuracy. Thereafter, a

bid value prediction system has been implemented to predict bid values of players. Regression techniques used for this. Finally, a mathematical equation formed to calculate the winnability of a team. This equation is a linear relationship between a team’s winnability and a weighted sum of players’ performances. Using this equation we can then assemble an optimal cricket team to enhance the winnability by using constraint optimization

techniques where the budget is the constraint.

• • Sanjeewa Kumara

  • Chamini Prashakthi

  • Sasitha Rajapaksha

  • Titus Nandakumara

In this paper, we examine how the Biofeedback can be used to improve user experience while playing the first person shooter game. Biofeedback is used to feed the body information of a real person to the game. Therefore, we are going to control and enhance a FPS game using some physiological functions of a human by mapping with the game character. We demonstrate the concept through a simple video game using two sensors to

detect the physiological states of the real player. Those two sensor devices are called as OpenBCI, which catches the Electrocardiography (ECG) signal and the Electrooculography (EOG) signal and Galvanic skin response sensor, which capture the skin conductance. Using these measurements, we can check the player’s excitement, eye movement, and the tiredness at the moment. If the excitement level become higher and the tiredness

become lower, the speed of the player will be increase, targeting for an aim will be high and generating enemies per time will be increase. If the excitement level became lower and the tiredness become higher, all the previous results will be happen in opposite way. If player looks left side, the screen will rotate left side by 15 degrees and for right side screen will rotate right side by 15 degrees. The major aim of this project is leads players to feel as real life experience while playing. Moreover, the game become addictive when it has this kind of features. In addition, another goal of adding this feature to the game is to control our body ourselves.

• • Prageeth Wanigasekara

  • Subhani Munasinghe

  • Pavinaa Thavapalan

  • Mr.Malintha Adikari

In recent years, social media emerged as a powerful resource to improve the management of crisis situations such as disasters triggered by natural hazards. In this project we focus on disasters triggered by natural hazards like floods, tsunami and cyclones. The objective of the work is to develop a

system that can be employed in natural disaster management. Disaster management is the creation of plans through which communities reduce vulnerability to hazards and cope with disasters. A disaster management model, which implements social media data mining techniques, can help the house holders in protecting their lives and properties from severe damages, based on the data retrieved from social media like facebook and

twitter. Data mining is a powerful technology for the extraction of hidden predictive and actionable information from large databases that can be used to gain deep and novel insights. Using data mining techniques on social media, the area which is going to be affected by a disaster, the spreading direction of a disaster and some other very useful features can be predicted.

• • Vimukthi Perera

  • Ching Shi

  • Brian Udugama

  • Titus Nanda Kumara

A major research focus in automobile development is improvement of safety. The main cause for road accidents is the distractions to the driver. Most distractions are in the form of emotional changes that result in unfitting states of mind. Existing methods of detecting a sleepy driver using image processing are proven to be challenging in practice due to the variations in the lighting condition. Further, it is insufficient to detect sleepiness

and fatigue as there are several other emotional conditions which could cause a driver to be in an unfitting state for driving. Such states of the driver could be identified using basic parameters of an ECG. In this research, different patterns in the ECG of the driver and patterns in the motion of the vehicle were identified for each emotional state to predict the driver’s emotional condition and warn if it tends to unsafe driving. Patterns in the

motion of the vehicle were analyzed in terms of the vehicle speed and the change in the acceleration. A Heart and Brain SpikerShield was used to obtain the ECG of the driver and an MPU-6050 IMU was used to gather the acceleration data of the vehicle. Collected data is sent to an Android device via Bluetooth for further processing. We were able to recognize the changes in the ECG and the driving pattern of a drowsy driver and an

aggressive driver. Accuracy of the emotion detection was verified by comparing the results against known methods.

• • Lakshitha Madushan

  • Pahan Madusha

  • Darshi Dineshika

  • Hasindu Gamaarachchi

General Purpose Computing on graphics processing units (GPGPU) has enabled inexpensive high performance computing for general-purpose applications. Compute Unified Device Architecture (CUDA) is a programming model which provides a platform to exploit parallel computing power of GPU using C/C++ languages. CUDA Dynamic Parallelism (CDP) is a feature introduced in CUDA Kepler architecture, which enables a CUDA kernel to create and synchronize new nested work. Huge computational time for data mining processes is a significant challenge met by data scientists. One solution for the above is the use of massive parallelism enabled by GPUs in data mining processes. This is a project done to provide a CUDA library for programmers to efficiently run data mining tasks with the use of NVIDIA GPUs. The library provides three popular data mining algorithms, namely Apriori, K-means clustering and Random Forest classification. The library not only accelerates the data mining process, but also provides an API for developers which can be used easily without much knowledge in CUDA. The researchers also have attempted to improve performance of this library by using better data structures and CUDA Dynamic Parallelism (CDP). The researchers main focus has been on building a CUDA library for developers and improving the performance of Apriori

algorithm using data structures suitable for GPU computations and CUDA Dynamic Parallelism (CDP). The experimental results of this research show how CPU and GPU performance differ and how using data structures suitable for GPU computations and CUDA Dynamic Parallelism (CDP) give better performance for the algorithms in the library. Keywords – Graphics Processing Unit(GPU), CUDA, Data mining, Apriori algorithm, Random Forest algorithm, K-means algorithm, Dynamic Parallelism

• • Lashan Faliq

  • Himasha De silva

  • Dinuka Nadeeshan

Traditional farming methods are becoming ineffective due to the increase of population and reducing land for cultivation. Therefore precision agriculture or more precisely greenhouses are important in this day and age also to get the maximum precision in the controlled environment we discussed the importance of a controlling and monitoring system to be in effect. Further, we discussed why a Wireless Sensor Network or WSN was used and which WSN is most suited to a monitoring and control system. Further, it was discussed about most of the frequently used sensors and the system architectures used in most of the papers and how our architecture is similar or different from those. Then we discussed the uses of self-learning and image processing in the field of precision agriculture and how our project uses a deep learning model to control the actuators. Finally, we are using an image processing model to identify optimal conditions of specific plants in a Controlled Environment.

• • Nishara Kavindi

  • Peshali Randika

  • Prabashi Meddegoda

Emerging of infectious diseases such as Dengue, have become a major challenge for the world. Use of indicator-based surveillance systems is the traditional approach of monitoring diseases, which uses structured data. Use of event-based surveillance systems is the modern approach, where unstructured data such as information from the internet and social media is Used. In Sri Lanka, there are indicator-based systems established for detecting and monitoring Dengue occurrences. But it still remains a major health problem. Therefore testing and implementing the other approach to strengthen the traditional system is important. There are successful event-based systems implemented in other countries. But none of them gives detailed information about Dengue spreading in Sri Lanka. Our objective is to address this issue through an automated system which queries for newly published online news articles and classify them as Dengue-related or not, extract useful information out of Dengue-related articles about Dengue outbreaks in Sri Lanka, store them in a database and visualize through a web application. In this paper, we describe data acquisition, classification, data extraction, data storing and the visualization processes of the system.

• • Theekshana Dissanayake

  • Yasitha Rajapaksha

  • Heshan Sandeepa

Dynamic Difficulty Adjustment (DDA) in computer games is a relatively new research area which focuses on improving the gaming experience by adjusting the difficulty level of the game depending on user performance. This article focuses on the design and the implementation of a performance-based DDA system which employs a combination of a neural network and a fuzzy system. In this system, a multilayer perceptron (MLP) neural network act as the difficulty detector and a fuzzy system act as the difficulty adjuster. Both these components were integrated into a first-person shooter game developed from the scratch which has thirty sections. The MLP neural network developed uses the game parameters to predict the difficulty level of the next section, and then feed the results into the fuzzy engine. After that, the fuzzy engine incorporates the game state parameters and the difficulty value with the knowledge base and computes an adequate difficulty adjustment. Finally, an experiment was conducted to compare the capabilities of the DDA integrated game with the traditional game. Compared to the base game, the developed DDA integrated game improved the players ability to finish the game by 30% and the game reduced the number of failed attempts by 76%.

• • W.M.K.D. Abeysinghe

  • K.L.D. Deshapriya

  • W.A.M.N. Weerasooriya 3

Traditional learning methods such as reading books, pdf and course materials are being out of date and smart learning and adaptive learning is emerging with the growing technologies. The ultimate goal of this paper is to present some improvements to existing adaptive testing systems in terms of  Moodle Adaptive Quiz Plugin. The major improvement is to integrate question classification with the adaptive quiz plugin. Question classification is done based on subject areas and difficulty levels of the questions. For that data mining techniques have to be taken into consideration which includes the details of data preprocessing and comparison between different classification techniques. This paper presents the detailed implementation process, pros and cons and the comparison between Decision Tree, Naive Bayes, SVM (Support Vector Machines), Random Forest and Artificial Neural Network in terms of question classification based on subject areas. And also the data preprocessing step for acquiring the train data set and accuracy scores of the above classifiers have also being descriptively presented. The classification of the questions from the difficulty levels also can be stated as another phase of this paper. A pre-answered question scheme was separated regarding difficulty levels using their accuracy scores. These questions and difficulty level classification also are done using previously mentioned classifiers and their results also included in this paper. The results of this classification could be integrated with the adaptive quiz plugin as future work. Moreover, some improvements for the existing Adaptive Quiz Plugin has also suggested in the context of this paper.