Portfolio
-
Exploring Computers Dataset With R
This project was developed during the mini course "Data Science with R" given by Professor Doctor Paolo Moser at the University of the State of Santa Catarina. In this course we carry out an exploratory analysis of a dataset with information about computers, the ultimate goal being supervised learning to determine the price of the computer according to its characteristics.
-
Bay Area Bike Share Review
This project was carried out in the Data Science I course given by Udacity and aimed to carry out an exploratory analysis of data from a bicycle rental company. It was possible to draw several insights to better understand the business, how the weather affects the rental of bicycles or even what is the best and worst schedule for bicycle rentals.
-
Revealing wiki4HE Data
This project was carried out as a technical test for the vacancy of data scientist at DTI Digital. The objective was to carry out the analysis of the questions regarding Perceived Enjoyment. For that, exploratory data analysis techniques were used and several groups of teachers and their answers to these questions were verified. Thus, other ideas about the data were drawn.
-
A Dynamic Risk Assessment System
In this project the company asked to create, deploy, and monitor a risk assessment ML model that will estimate the attrition risk of each of the company's 10,000 clients. If the model you create and deploy is accurate, it will enable the client managers to contact the clients with the highest risk and avoid losing clients and revenue.
-
Census Bureal Classification
This project is part of Unit 4: Deploying a Scalable ML Pipeline in Production. The problem is to build a machine learning application that predicts an employer's annual income more than $50K using the census income dataset from UCI. The application is deployed using FastAPI, with CI and CD using Github Actions and Heroku respectively.
-
Build an ML Pipeline for Short-Term Rental Prices in NYC
In this project we are working for a property management company renting rooms and properties for short periods of time on various rental platforms. So we need to estimate the typical price for a given property based on the price of similar properties. The company receives new data in bulk every week. The model needs to be retrained with the same cadence, necessitating an end-to-end pipeline that can be reused.
-
Predict Customer Churn
The Project is aimed at identifying credit card customers who are most likely to churn. Machine learning algorithms have been used to achieve this objective. PEP8 standard has been followed while writing the Python code. Python Codes could be found in the repository along with the testing and logging python script.
-
Finding Donors For Charity
In this project it was predetermined that people who earn more than 50 thousand annually were more likely to make donations to charity. Thus, through a database containing characteristics of people who earned above and below 50 thousand, it was asked to create a model capable of determining whether the person earns more or less than 50 thousand per year according to their characteristics.
-
Analyzing Stocks With Machine Learning
In order to learn more about the financial market, this machine learning course for finance was held. It was possible to learn about how to analyze stocks separately and how to analyze the performance of stock portfolios. Regression, visualization techniques, bollinger bands and moving averages were used.
-
Using News To Predict Stock Movement
This project was completed as a conclusion of the Machine Learning Engineer course, from Udacity. However, it was chosen because it is a Kaggle competition. The objective was to develop a model to predict whether the market would rise or fall according to news labeled as good or bad. The results were not the best, but several concepts were studied in this project.
-
Creating Customer Segments
As a final project for the unsupervised learning section of the Machine Learning Engineer course, it was proposed to segment customers and reduce the dimensionality in standardized data on the quantity of products sold in each category. PCA and clustering techniques were used to analyze the data.
-
Unravealing Real Estate Ads
In this project proposed by the company Data Sprints as a technical test, data received by the company about real estate advertisements were received. However, the data came with some problems like missing and repeated data. It was proposed to perform a dimensionality reduction and analyze the main component of the data. In addition, pre-processing was performed to eliminate missing data.
-
Teach A Quadcopter How To Fly
Final project of the section about reinforcement learning on the machine learning engineer course by Udacity. It was proposed to use techniques such as Q-learning to make a quadcopter learn to take flight.
-
Monte Carlo Methods To Play Blackjack
Reinforcement learning module project developed to train an agent to play blackjack using Monte Carlo methods.
-
Temporal Difference For Cliff Walking
Reinforcement learning module project developed to train an agent to learn how to find the best path to the goal without falling off the cliff, using the time difference methods.
-
Dynamic Programming For Frozen Lake
Reinforcement learning module project developed to train an agent to learn how to find the best path in an environment where each path is associated with a probability (the agent can slip into the lake), using dynamic programming.
-
Getting Started With NLP
Two notebooks were developed to study the modules used in natural language processing (NLP), word2vec which makes a vectorization of textual words and a learning process using the dataset not mnist.
-
Getting Started With Tensorflow
Course acquired at udemy, presents the main architectures of deep neural networks such as CNN, LSTM, RNN, GANS and others.
-
Character Recognition Project
This project carried out in the discipline of computational intelligence during the graduation in software engineering proposed the creation of a model capable of verifying handwritten digits (dataset semeion). In addition, the model was put into production in an android application using the tensorflow lite.
-
Smart Searches In The Sokoban
The Japanese game sokoban aims to move the boxes to the corresponding positions, using a character to push the boxes without letting them get stuck. For that, several search methods (width, depth and others) were developed, to find the solution for some maps proposed by the professor in the discipline of Computational Intelligence during the graduation in software engineering.
-
Max Clique Vertex Weighted Problem
This project was carried out in the subject of Project Integrator three, during the Software Engineering course held at UDESC. The objective was to find a click with a greater sum of the weights within a graph. That is, a click is a sub-graph in which all vertices are connected by an edge. Several types of heuristic searches were generated for the problem, such as Iterated Local Search.