Research on Wil's blog

A Thorough Evaluation of Demand Prediction Models: Machine Learning, Deep Learning, and Statistical Techniques for Import Businesses

Fri, 17 Oct 2025 00:00:00 +0000

Springer Nature / Emerging Trends in Information Systems and Technologies: WorldCIST 2025 Volume 5 · Oct 17, 2025

Nowadays, managing demand in companies is crucial to avoid storage overcosts, stockouts and to improve the service level of companies. To address this scenario, demand predictions through models and algorithms emerge. Therefore, this research aims to evaluate the performance of seven prediction techniques applying machine learning, deep learning, and statistical methods. To validate our experiments, we used Dickey–Fuller, Shapiro–Wilk, Friedman, and Wilcoxon post-hoc statistical tests on the predictions of the models using demand records from a Peruvian import company. The results indicated that deep learning and statistical models have significantly better predictions than machine learning models. In particular, the LSTM, CNN, ARIMA, and Holt-Winters models significantly improve accuracy compared to the Ridge Regression, Random Forest Regressor, and Decision Tree Regressor models. Compared to machine learning models, statistical and deep learning models improve accuracy in a range from 66.01 to 86.10%. These results highlight the statistical advantage of deep learning and statistical models in demand prediction, with the LSTM model showing the lowest error.

An extended study of a cloud-based recommender system for competitive programming platforms with machine and deep learning

Fri, 01 Nov 2024 00:00:00 +0000

Brazilian Journal of Development · Nov 1, 2024

Show publication

Nowadays, online judges are very important to improve programming skills for education and technology companies. For this reason, there are many online judges that include large sets of programming challenges. This creates an information overload problem that affects students due to their lack of expertise in choosing the correct challenge to solve, resulting in frustration and a loss of interest in this topic. To solve this scenario, recommender systems appear, but programming judges have not delved much into it. Consequently, this research aims to evaluate the performance of six selected collaborative filtering techniques via a cloud-based software architecture. To validate our experiments we used real online programming judges like CodeChef and NinjaCoding using cloud based architecture with Amazon Web Services, evaluated through Friedman and Wilcoxon statistical tests. The results indicated that Singular Value Decomposition is the best model evaluated with RMSE metric and the fastest in execution time with big datasets.

Sistema de recomendación basado en modelos híbridos de filtrado colaborativo para jueces de programación en línea

Mon, 01 Jan 2024 00:00:00 +0000

Tesis de maestría, Universidad Nacional Mayor de San Marcos · 2024

Show publication

Hoy en día, los jueces en línea de programación desempeñan un papel fundamental en ayudar a los estudiantes a desarrollar habilidades de programación que les son útiles tanto en empresas tecnológicas como en la educación. Debido a esto, muchos jueces en línea ofrecen una cantidad considerable de retos de programación; sin embargo, esta sobrecarga de retos puede frustrar a los estudiantes y hacer que pierdan interés en resolverlos, especialmente debido a su inexperiencia a la hora de seleccionar el próximo desafío de programación. Los sistemas de recomendación surgen para abordar esta situación, aunque en el contexto de los jueces de programación no han sido estudiados exhaustivamente. Por lo tanto, el objetivo de este estudio es mejorar los sistemas de recomendación de jueces de programación mediante modelos híbridos de filtrado colaborativo, utilizando ensamblado vía stacking con el modelo meta de RandomForest y Optimización Bayesiana. Para validar nuestros experimentos, utilizamos los jueces de programación en línea NinjaCoding y CodeChef. Los resultados se evaluaron mediante las pruebas estadísticas de Friedman y Wilcoxon, las cuales indican mejoras entre el 26,08% y 58,13% en la predicción de retos en los jueces CodeChef y NinjaCoding, respectivamente.

Content-Based Recommendation System for Programming Judges using Natural Language Processing and Deep Learning

Tue, 26 Sep 2023 00:00:00 +0000

RPCS · Sep 26, 2023

Show publication

In the field of education and technology companies, online judges play an important role in the development of programming skills because on these platforms students must solve challenges using specific programming languages. However, the sheer number of coding challenges available can be overwhelming for students, leading to frustration and loss of interest. To resolve this situation, re-commender systems can be an effective solution. However, programming judges have not delved far enough into this area. Therefore, this research focused on evaluating six artificial intelligence techniques through a cloud-based architecture for the prediction of the level of difficulty from the statements of the problems to be coupled to a recommendation system. To validate the experiments, a real CodeChef programming judge was used and the experiments were evaluated through statistical tests. The results indicated that the BERT model is the best for predicting the level of the problems, which helps the recommendation system to improve the learning experience of the students in the online programming judges.

A Cloud Based Recommender System for Competitive Programming Platforms with Machine and Deep Learning

Wed, 23 Aug 2023 00:00:00 +0000

VIII Congresso sobre Tecnologias na Educação · Aug 23, 2023

Show publication

Hoje em dia, os juízes online são muito importantes para melhorar as habilidades de programação para empresas de educação e tecnologia. Por esse motivo, existem muitos juízes online que incluem grandes conjuntos de desafios de programação. Isso cria um problema de sobrecarga de informações que afeta os alunos devido à falta de experiência em escolher o desafio correto para resolver, resultando em frustração e perda de interesse por esse tópico. Para resolver esse cenário, surgiram os sistemas de recomendação, mas os juízes de programação não se aprofundaram muito nisso. Consequentemente, esta pesquisa visa avaliar o desempenho de seis técnicas de filtragem colaborativa selecionadas por meio de uma arquitetura de software baseada em nuvem. Para validar nossos experimentos, usamos juízes de programação online reais como CodeChef e NinjaCoding usando arquitetura baseada em nuvem com Amazon Web Services, avaliados por meio de testes estatísticos de Friedman e Wilcoxon. Os resultados indicaram que a Singular Value Decomposition é o melhor modelo avaliado com a métrica RMSE e o mais rápido em tempo de execução com grandes conjuntos de dados.

Evaluation of source code in ACM ICPC style programming and training competitions

Mon, 01 Jan 2018 00:00:00 +0000

Avances en Ingenieria de Software a Nivel Iberoamericano, CIbSE 2018 · 2018

Show publication

The execution of potentially dangerous source code is one of the main problems in the platforms of programming contest. In this research we propose NinjaCoding, which is a web platform to evaluate the source code in programming contests and training ACM ICPC style. NinjaCoding is designed to be a scalable, safe and economical system, in addition our proposal ensures the execution of source code through the use of containers and subprocesses. Students can use NinjaCoding to participate in past or active contests and in training problems with live feedback. This system has been tested in several programming contestss that have taken place in various congresses and universities in different cities of Peru, qualifying more than 1000 presentations in the languages C/C++, Python and Java. This document describes and discusses the architecture and the implementation that is used in NinjaCoding.