Sentiment Analysis of Amazon Product Reviews using LSTM
The goal of this project was to develop an automated machine learning solution that could accurately categorize Amazon reviews as Positive, Neutral, or Negative, going beyond the limitations of traditional star ratings. The intent was to provide vendors with specific insights from reviews to enhance product quality and user experience on Amazon.
Tools such as Python, Natural Language Processing (NLP) libraries, and Long Short-Term Memory (LSTM) neural networks were employed to create the model.
The outcome was an efficient model that could identify and categorize negative reviews with a high recall of 84%. This underscores the significance of the Recall metric in capturing the majority of negative reviews. If implemented, Amazon could save an estimated $15,330,000 annually!
The tools used were:
- Python, Pandas, Keras, Tensorflow
- Jupyter Notebook
- Git,Github
- Long Short-Term Memory Neural Networks(LSTM)
Classification of Brain Tumors using Deep Learning.
The aim of the project was to build a classifier that would increase the accuracy of brain tumor diagnoses at the Health Care Hospital.
Tools such as Python, Keras, Tensorflow, and Convolutional Neural Networks (specifically the VGG-16 model) were used to build the model.
The result was a machine learning model with 99% accuracy in correctly classifying brain tumor images.
The tools used were:
- Python, Pandas, NLP libraries, Tensorflow, Keras
- Jupyter Notebook
- Git,Github
- Convolution Neural Networks(CNN)
A/B Testing for a Mobile Game.
The objective of this project was to design an A/B test for a Mobile Game. As players progress in the game, they encounter gates that force them to wait a while before they can proceed or make an in-app purchase. In this project, we will analyze the results of an A/B test in which the first gate in Cookie Cats was moved from level 30 to level 40. Specifically, we will analyze the impact on player retention.
Tools such as Python, Data Visualization libraries, and Statistical concepts were used to understand, model, and apply the A/B test.
The result concluded that there was a significant effect in moving the gate to level 40. However, this effect would lead to a decrease in revenue, making it more sensible to keep the gate as it was at level 30.
The tools used were:
- Python, Pandas, Matplotlib, Seaborn
- Jupyter Notebook
- Git,Github
- Statistics(Teste Chi-Squared)
Building a Loyalty Program with Customer Clustering.
The objective of this project was to develop a loyalty program to segment the customers of a fictitious E-Commerce company from the United Kingdom using clustering techniques.
Tools such as Python, Statistics, and unsupervised Machine Learning techniques were used to segment a group of customers based on their purchase characteristics. With these characteristics, a Dashboard was built in Metabase to monitor the metrics built in the project.
In addition, I used Software Engineering techniques to build a Cloud infrastructure on Amazon AWS to automate the production process.
The result of this solution, if implemented, would result in an increase in the company's revenue by about 12.5%.
The tools used were:
- Python, Pandas, Matplotlib, Seaborn
- Jupyter Notebook
- K-Means, Hierachical Clustering, GaussianMixedModel
- AWS Cloud(EC2, S3, Postgres, SQLite)
- Git,Github
- Estatística
- Metabase Visualization
Health Insurance Cross-Selling
The goal of this project was to help a health insurance company also sell car insurance to its existing customers, i.e., to cross-sell. For this purpose, a Machine Learning model was developed to rank customers most likely to purchase car insurance, facilitating the company's communication strategy, optimizing its business, and reducing costs.
Tools like Python programming for Exploratory Analysis and Data Manipulation, Business Insights through Statistics, web tools like Github, and cloud servers like Heroku were used. I also used Machine Learning algorithms, specifically Supervised Classification algorithms.
The result was a spreadsheet in Google Sheets that, given a customer base, made predictions through an API that fetched the Machine Learning model from the Cloud and returned this customer base sorted by the highest probability of purchasing car insurance.
The tools used were:
- Python
- Jupyter Notebook
- Heroku Cloud
- Machine Learning
- Git, Github
- Statistics
Sales Forecast for a Pharmacy Chain.
In this project, a sales forecast was made for the Rossmann pharmacy chain for the next 6 weeks, which can be accessed by a bot on Telegram from any internet-enabled device.
Tools like Python programming for Exploratory Analysis and Data Manipulation, web tools like Github, and ORender were used. Machine Learning concepts and models such as XGBoost, Random Forest, among others, were also utilized.
The final result was a Telegram Bot that accesses the Machine Learning model published in the Cloud and provides information about the sales forecast for a specific store.
The tools used were:
- Python
- Jupyter Notebook
- Render Cloud
- Machine Learning
- Github
Development of a Dashboard for a Restaurant Marketplace using Streamlit.
In this project, using Python programming concepts, Data Manipulation, Strategic Thinking, and Business Logic, along with web development tools like Streamlit and Github, a Management Panel was developed with the main business metrics for a restaurant marketplace operating in several countries and continents.
The final result was a Dashboard hosted in a Cloud environment available through the web. The panel can be accessed using any internet-connected device.
The tools used were:
- Python
- Jupyter Lab
- Terminal
- Streamlit
- Streamlit Cloud
- Github
