International Conference on Advances in Computing and Technology (ICACT)
Permanent URI for this communityhttp://repository.kln.ac.lk/handle/123456789/15607
Browse
3 results
Search Results
Item Stock Market Analysis and Prediction.(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2017) Shakya, A.; Pokhrel, A.; Bhattarai, A.; Sitikhu, P.; Shakya, S.Stock price and stock index price forecasting system, used by investors and financial managers to describe the market and compare the return on specific investments, has been a topic of research for very long now. When in the stock market, there are more buyers than there are sellers, the price must adapt or no trades are made. This tends to drive the price upwards, increasing the market quotation at which investors can sell their shares, enticing investors who had previously not been interested in selling and vice versa. These demands and supplies are ever changing, resulting in highly-fluctuating, non-linear stock prices which poses a threat against the credibility of those prediction systems which only view the market from one perspective. For a reliable system, it is therefore important to explore the market on multiple grounds, basically through Technical, Fundamental and News Analysis. Under Technical Analysis, SMA (Simple Moving Average) is used as a preliminary data smoothing technique, which helps reduce the fluctuations substantially. Artificial Neural Networks (ANNs) is then employed to analyze the nonlinear relationships between the stock closed price and various technical indexes, and to capture the knowledge of trading signals that are hidden in historical data. Features like traded share, traded volume, opening price, closing price, high price and low price are fed as an input parameter in Neural Network. Backpropagation algorithm is then implemented to train the given Network model. The neural network layers and neuron numbers in hidden layers are then tuned by training and validating the data set iteratively. The fundamental analysis involves thorough study of financial statements of companies, also known as quantitative analysis. This involves looking at assets, liabilities, revenue, expenses and all other financial aspects of a company. It gives insight on the company's future performance. The results moreover reflect the company's success or failure over the long term than immediate future. Hence, unlike technical analysis, it helps predicting stock price on a long run. In news analysis, we focus on understanding the news sentiment and its affects which may cause the investors to either buy or sell the shares based on positivity or negativity of the news. The news analysis problem can be mapped into similarity based classification. A set of vectors are created from analysis of historical news, where each component of a vector represents the features in data set. The required labeling are done based on historical rise/fall of stock prices. The increase or decrease of the trend is then predicted based on similarity measures. In short, news analysis predicts the price of share of the following day by comparing the most recent news with past news using Knearest neighbor algorithm. Thus, through the circumstantial application of the above-mentioned analysis, the paper proposes to predict the stock market in a more generalized manner.Item Artificial Neural Network based Emotions Recognition System for Tamil Speech.(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2017) Paranthaman, D.; Thirukumaran, S.Emotion has become the important part in communication between human and machine, so the detection of emotions has become important part in pattern recognition through Artificial Neural Network (ANN). Human's emotions can be detected based on the physiological measurements, facial expressions and speech. Since human shows different expressions for a particular emotion when they are speaking therefore the emotions can be quantified. The English speech dataset is provided with descriptions of each emotional context available in Emotional Prosody Speech and Transcripts in the Linguistic Data Consortium (LDC). The main objective of this project describes the ANN based approach for Tamil speech emotions recognition by analyzing four basic emotions sad, angry, happy and neutral using the mid-term features. Tamil speeches are recorded with four emotions by males and females using the software “Cubase”. The time duration is set to three seconds with the sampling frequency of 44.1 kHz as it is the logical and default choice for most digital audio material. For the simulations, these recorded speech samples are categorized into different datasets and 40 samples are included in each dataset. Preprocessing includes sampling, normalization and segmentation and is performed on the speech signals. In the sampling process the analog signals are converted into digital signals then each speech sentence is normalized to ensure that all the sentences are in the same volume range. Next, the signals are separated into frames in the segmentation process. Then, the mid-term features such as speech rate, energy, pitch and Mel Frequency Cepstral Coefficients (MFCC) are extracted from the speech signals. Mean and Variance values are calculated from the extracted features. To create the classifier for the emotions, the above statistical results as an input matrix with their related emotions-target matrix are fed to train, validate and test. The neural network back propagation algorithm is executed by the classifier to recognize completely new samples of Tamil speech datasets. Each of the datasets consists of different combinations of speech sentences with different emotions. Then, the new speech samples are assigned to identify the recognition rate of the speech emotions using the confusion matrix. In conclusion, the selected mid-term features of Tamil speech signals classify the four emotions with the overall accuracy of 83.45%. Thus, the mid-term features selected are proven to be the good representations of emotions for Tamil speech signals and correctly recognize the Tamil speech emotions using ANN. The input gathered by a group of experienced drama artists who have the voice with the good emotional feelings would help to increase the accuracy of the dataset.Item Optimizing the Member Selection for Ensembles of Classifiers: An Application of Rainfall Forecasting in Sri Lanka(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka, 2016) Nagahamulla, H.R.K.A collection of classifiers trained to do the same task is called an ensemble of classifiers. Ensembles can be created using a set of classifiers of the same type or using a set of classifiers in different types (Artificial Neural Networks (ANN), Support Vector Machines (SVM), Decision Trees, ect.). The generalization ability of an ensemble is significantly increased than that of a single classifier. To achieve increased generalization ability the members of an ensemble has to be accurate (able to produce correct forecast) and diverse (errors in different regions of the error space). However accuracy and diversity are two conflicting conditions that have to be balanced carefully to achieve good performance. Thus members for an ensemble need to be selected carefully in order for them to have the perfect balance between accuracy and diversity. This study aims to optimize the member selection for the ensembles using Genetic Algorithms (GA) to increase the ensemble performance in the context of time series forecasting. The selected application is rainfall forecasting in Sri Lanka. Rainfall is very difficult to forecast accurately because it is a very complex hydrological process. Forecasting rainfall requires manipulating huge datasets with large number of variables. But accurate rainfall forecasts are in high demand because of the close relationship rainfall has with human life. There are three steps in creating an ensemble; creating the pool of classifiers, selecting the members for the ensemble from the pool and combining the selected members using a combiner method. The performance of the ensemble depends on the techniques used in each of these steps. First a pool of classifiers, including different types of classifiers such as SVM, Back Propagation Network (BPN), Radial Basis Function Network (RBFN) and Generalized Regression Neural Network (GRNN) was created by training the classifiers using different training data. Then a number of ensembles were created by selecting different combinations of classifiers from the pool randomly and combining them using a separate GRNN. These ensembles were the initial population of the GA. A simple binary genetic algorithm was then used to create new generations of ensembles and find the ensemble that gave the best result. The fitness of the ensembles were calculated to balance the accuracy and the diversity of the ensemble. The chromosomes were ranked and sorted according to their fitness. Then, the mating pool was prepared by selecting the chromosomes with highest fitness and the pairs were selected using roulette wheel rank weighting. Mating took place using one point crossover with 0.6 crossover probability and the new generation was mutated with 0.1 mutation probability. To train and test the models rainfall data from 1961 to 2001 (41 years) of Colombo, Sri Lanka is used. Input data set consisted of 26 variables obtained from the NCEP_1961-2001 dataset and the output data was daily rainfall of Colombo. The dataset was partitioned to training data (first 60%), validation data (next 20%) and testing data (the remaining, more recent 20%). To create different training datasets from the available training data moving block bootstrap method was used. The dataset containing 10958 records was split into 9863 overlapping blocks of length 1096 and out of these 9863 blocks 10 blocks were selected to train each classifier. To validate the proposed method another two ensembles were created using two well known ensemble creation methods bagging and boosting. The performance of the best ensemble (ENN-GA) was compared with the performance of a single SVM, BPN, RBFN, GRNN, the best performing ensemble in the initial population (ENN), bagging model and the boosting model. Forecasting accuracy of each model was measured for the test dataset using Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and the Coefficient of Determination (R2). The best performing ensemble comprised of two SVM, three BPN, two RBFN and five GRNN. The number of generations for convergence was 287. The following table summarizes the results for individual classifiers, ENN, bagging, boosting and ENN-GA.The proposed model ENN-GA gave more accurate results than the single classifiers used in the study with smaller RMSE and MAE values and larger R2 and the time and space requirements were very small. The proposed model managed to predict the overall rainfall with reasonable accuracy; zero rainfall accurately, smaller rainfall with slight differences and some higher rainfall with considerable differences. These higher differences were obtained for very high rainfall that occurred suddenly. Although the number of these occurrences were very few the difference between the actual and forecasted rainfall was high. The RMSE values were larger compared to MAE values because the errors in high rainfall were magnified in RMSE. The proposed method outperform the single classifiers, ENN model and bagging and boosting models in forecasting rainfall for Colombo, Sri Lanka.