Browsing by Author "Liyanage, C.R."

Now showing 1 - 3 of 3
  • No Thumbnail Available
    Item
    Predict Human Personality based on Handwritten Signature
    (Uva Wellassa University of Sri Lanka, 2021) Tharuka, R.M.S.; Liyanage, C.R.
    Personality is a unique thing that everyone has and it shows how a person acts both in daily life and at work. Therefore, tracking a person's personality has become more important, especially for an employer. Within this context, the purpose of this research is to identify a person's personality through big five-factor personality traits based on his/her handwritten signature. The majority of earlier researchers have focused on analyzing handwritten signatures to describe personality with the help of graphology. The current research was designed a way to apply graphology on the signature image and improve the performance using neural networks. In this study, the personality of a person was evaluated based on four selected features of a signature -namely, the size, curved start, pen pressure, and underline. Further, an online questionnaire, which was conducted with the participation of 500 selected individuals, has been utilized to measure and gather the personality of each person. The complete system evaluates signature samples based on the above features and divided into four modules. Then these four modules were fed into the feature extraction model, which analyzed the input image with the Convolutional Neural Network (CNN) model and all four features were extracted from the signature data set. After that, the extracted features were combined with the online questionnaire test result to help with supervised learning. As the final output, this model predicts the correct big five-factor personality values with 85% accuracy, when a person wrote his/her signature on a paper. This solution is unique as this predicts the big-five factor personality traits based on the signature for the first time and this is a more efficient approach compared to other existing work Keywords: Convolutional Neural Network; Personality Traits; Signature Analysis; Supervised Learning
  • No Thumbnail Available
    Item
    A Supervised Learning Approach to Detect Black Pepper Adulteration
    (Uva Wellassa University of Sri Lanka, 2021) Hansani, N.N.; Liyanage, C.R.
    Black pepper is one of the widely planted spices in the world and its cost is very high when compared to other spices. Many People earn huge income from black pepper by exporting and selling. Therefore, they mix unwanted adulterants, such as stones, weeds and other low-cost items in order to increase the quantity and the profit. Out of these adulterants, papaya seeds are very common, as their appearance is very similar to black pepper seeds. Those malpractices will reduce the quality of pepper samples and it is difficult to control as these two types of seeds cannot be easily classified even by an expert eye due to the smaller size in bulk samples. Hence, more advanced solutions are required to determine the adulteration of pepper samples. Currently, there are some existing studies and methods; one is a manual method to separate papaya and pepper seeds using water by considering their weights. Further, there are chemical methods, such as the Thin-Layer Chromatography (TLC) approach to detect the adulterated papaya seeds by using mixed samples of black paper powder and ground papaya seed. In this study, a classification method was proposed to differentiate black pepper from papaya seeds using the Convolutional Neural Network (CNN) technique. The images of the samples were captured using a high-resolution digital camera and the features, such as size and shape were extracted to classify the seeds. Next, these features were fed as the input to the CNN model for the classification task. The experimented model was able to successfully label the papaya and black pepper seeds with an accuracy rate of 85.94%. As the main output of the model, the percentages of papaya and pepper seeds in the given sample were presented. To improve the accuracy of the model, high-quality images and more features such as texture and color will be used in future work. Keywords: Black Pepper Adulteration; Feature Extraction; Image Processing; Supervised Learning
  • No Thumbnail Available
    Item
    Utilization of Text-based Emotions in social media for Depression Analysis
    (Uva Wellassa University of Sri Lanka, 2021) Gedara, C.K.R.; Liyanage, C.R.
    As estimated, millions of people around the world suffer from depression every year. It is a common and serious medical illness that negatively affects the way a person feels, thinks, and acts. Emotions can be used to identify depressed people through changes in their moods, expressions of thoughts, ideas, and opinions. With the expansion of online social media networks, many people share their thoughts and opinions as text-based posts and these are rich sources of human emotions. The current study has used text posts and comments which were published on public groups on Facebook related to depression as the labeled dataset for text analysis. Although previous studies have detected depression using different techniques, such as facial expressions analysis, behavioral analysis, and investigation of linguistic characteristics of written text, the results were not adequate for life-saving applications. Therefore, intending to solve with higher accuracy, this study investigated a novel emotional intensity-based approach to detect depressed people. During this study, depression is identified by analyzing emotional intensities in-text sources using a supervised learning approach. The existing NRC lexicon model which contains 8 basic emotions; anticipation, trust, joy, fear, sadness, surprise, anger, and disgust was used to get similar word lists with real-valued scores of weights for each word in eight basic emotions. After pre-processing the text, the techniques, such as developing a Bag of Words to collect all the words related to emotions, performing vectorization to extract words from posts that are similar to the Bag of Words, and finding intensities for each emotion using cosine similarity were conducted. Finally, the intensities of all 8 emotions of each post were fed into a Feed-Forward Neural Network to learn and predict the pattern of intensities to classify depression or non-depression according to the text posts. As the main output of the study, people with depression were identified, with 90% accuracy according to the patterns of emotional words in their posts on social media. Keywords: Bag of Words; Cosine Similarity; Emotional Intensities; Feed Forward Neural Network; NRC Lexicon; Vectorization