Sergio Chavez

Inspired by the anime game, I successfully developed and optimised a neural network with Keras that classifies first-generation Pokemon images. I conducted the project with a learning approach, exploring simple to more complex approaches using Transfer Learning (DenseNet) and hyperparameter optimisation.

Using a database of over 1 million tracks, I developed a model to predict the popularity of Spotify songs. The project comprises a comprehensive comparison of supervised models such as Random Forest, Multilogistic Regression, Boosting & ANN.

Using the nutritional information of the products offered by McDonalds, explore the optimal clustering using and comparing different algorithms such as KMeans, Kernel KM, Hierarchical Clustering & EM. In addition, the processing includes PCA for dimensionality reduction.

Using Python, I developed and optimised a model focused on forecasting the sales of about 1000 shops of the Rossman chain in Germany. Among other things, the accuracy of the model was very high due to good decisions in the cleaning process.

(Project written in Spanish) Using R, in this project I explored the influence of educational level and wealth on the probability of having obesity for men and women in Peru in 2022. The analysis contemplates the exploration of data at the departmental level. Winner of the research incentive award from the Pontificia Universidad Catolica del Peru (#1 QS Ranking in Peru).

Using the Our World in Data database, I wrote SQL queries that answered 12 research questions related to the management and impact of the pandemic in over 150 countries. Part of the information collected was used to build a dashboard in Tableau (see next project).

Using my own queries, I created an interactive dashboard on the impact and management of the pandemic in over 150 countries. It covers data from January 2020 to April 2023.

I developed a preliminary model to predict the number of cumulative COVID19 cases for Peru. The project can be updated, optimized and modified to predict the situation in any country.

I faced the challenge of classifying more than 50,000 low quality images from the CIFAR100 database. I compared different approaches to the problem and built a neural network with a satisfactory level of accuracy. The project relies on a pre-trained neural network: MobileNet. Access to report here

(Group Work) Together with the team of master students (BeePy) we created, optimised and compared three statistical models to recommend service providers to the ACME company. This project was part of the Introduction to Python course at the University of Manchester and scored one of the top marks in the class.

(Project written in Spanish) In this project, I answer some interesting questions about one of the biggest events in sports. I found fascinating trends using very helpful libraries of R such as tidyverse, ggplot2, plotly and kable.

This was my first project using Python. It was an enjoyable challenge to make my first steps in this programming language meanwhile I answered some of the questions I had always had as a Pokemon fan