UPLB ICS Peak One

This paper details the design and implementation of TraveLBetter, a mobile application for public transportation at the University of the Philippines - Los Banos (UPLB). TraveLBetter enhances interactions between students, jeepney drivers, and the University Police Force (UPF). TraveLBetter allows students to queue online for jeepneys servicing the Forestry or Rural routes, enables drivers to view queued and petitioned users, and helps UPF monitor and track jeepneys within the campus. A Tailored Likert Scale Survey reveals that passengers experienced reduced perceived waiting time, drivers benefited from increased trip opportunities, and UPF valued the tracking ability.

College students are vulnerable to making poor dietary decisions and have a variety of food restrictions. ElbiEats a food recommender system that suggests dishes based on their nutritional value. Using a food’s nutritional composition, as listed on the Philippine Food Composition Table, the system acts as a dish review website and tracks a user’s nutrient intake against the Recommended Dietary Allowance. With the help of this system, restaurant owners can generate the nutritional composition of their dishes. Using a System Usability Scale, 48 respondents gave the system an average score of 82.5%, earning it the ”Excellent” adjective rating. However, the respondents recommended improvements to the system’s user interface and loading speed, and provided positive feedback and suggestions for additional features.

The Internet has an abundant amount of textual information which causes moviegoers to check out movie forums and websites for reviews before deciding to watch a movie. Thus, this study was written to develop a mobile application, Emofy, that retrieves movie reviews from Letterboxd and classifies them based on seven emotions – joy, sadness, anger, fear, love, surprise, and neutral – using a Multinomial Na¨ıve Bayes classifier. The application displays a pie chart that shows the emotional distri- bution of the movie reviews, generates a word cloud that shows the most frequently used words in the reviews, and lets the users browse the movie reviews based on emotion. The classifier was trained using a balanced dataset, split into two parts (70%:30%) for training and testing respectively. It yielded an 80% accuracy using a TF-IDF vectorizer with default parameters. Lastly, the mobile application was tested using the System Usability Scale (SUS) among twenty respondents and produced an average SUS score of 96.5.

The Laguna Youth Organizations Hub is a web application developed with the aim of amplifying volunteerism and volunteer-organization involvement. Several functionalities were laid out for various users of the Hub such as promoting upcoming activities, presenting organization advocacies, and joining in various youth-led activities and collaborating with organizations. Developed using T3 Tech Stack, the Hub garnered an average SUS score of 86.33 from thirty-two respondents.

This study investigated the feasibility of using YOLOv8, a deep learning model, for eggplant crop and weed detection on resource-constrained devices specifically, Raspberry Pi. A custom dataset was created and used to train and evaluate YOLOv8 models iteratively with k-fold cross-validation and grid search for hyperparameter tuning. Evaluation using a hold-out test set confirmed the effectiveness of k-fold cross-validation and demonstrated the potential benefit of using larger images (320x320) for accuracy (mAP@50 score). However, deploying a model on a microcontroller has to consider the trade-off between accuracy and inference speed. The smaller model (224x224) achieved faster inference (105ms, 8 FPS) but with lower accuracy (mAP@50 = 0.721). The larger model (320x320) achieved higher accuracy (mAP@50 = 0.794) but at a slower inference speed (220ms, 4 FPS). These findings provide valuable insights for further development and optimization of weed detection software for Project SASHA, considering the balance between accuracy and real-time performance on resource-limited devices.