https://hdl.handle.net/20.500.14353/512 2026-04-20T11:25:04Z https://hdl.handle.net/20.500.14353/448 Differentiation between organic and non-organic green onions using image classification with hyperparameter tuning Dela Cruz, Nerilou B. Differentiation between agricultural organic and non-organic crops involves professional laboratory techniques using expensive devices. This research domain requires a real-world dataset (RWD) which is limited depending on the subject or issue of the research study. Thus, this work presented real-world green onions image datasets collected from various locations in Iloilo, Philippines. The gathered datasets fit ground truth criteria with notable information (e.g., size, width, height, resolutions, the weather during the time it captures, and place) for similarity differentiation. Moreover, this study aimed to design and develop a non-intrusive image classification using Deep Learning (DL) methods such as Convolutional Neural Networks (CNN) and Transfer Learning Models provided hyperparameter tuning. Hyperparameters are sets of variables that govern the training process of DL models. These variables remained constant over the training process and directly impacted the performance until it acquired results around 99% training and 96.25% validation accuracies. With this, an application was developed and successfully assisted users in differentiating organic and non-organic green onions using image classification. 2022-07-01T00:00:00Z https://hdl.handle.net/20.500.14353/146 Classifier model based on text mining for discovery of depression on social media Deocampo, Nikie Jo E. Predicting suicidal people in social networks is a real social issue. Suicide due to depression or anxiety has always been a problem with strong socio-economic consequences. However, global provisions and services for identifying, supporting, and treating mental illness of this nature have been considered as insufficient. In this study, the researcher describes a complete process to dynamically collect suspected tweets according to a lexicon of topics persons with general depression and anxiety would usually post and talk about. The researcher provides a mechanism that automatically captures tweets indicating depression risk behaviors based on Sentiment Analysis and Natural Language Processing and each tweet is based on the polarity or sentiment weight of positive, negative or neutral. In building the model, Five Classification methods will be used for training and testing the data set. The Decision Tree algorithm provided the highest accuracy and precision which is followed closely by Naive Bayesian. The model with the highest accuracy will then be used to assist or support mental health test and decision support systems. The model was then tested and evaluated by a practicing psychologist using a same data set used for testing and training. 2019-04-01T00:00:00Z