My name is Luis, and I am currently studying Computer Science at UFPI. My main objective is to develop analyses and predictions using Machine Learning algorithms and programming tools, such as Python and SQL, to solve business problems and generate insights for companies. At the moment, I am a Machine Learning Engineer at SantoDigital.
Data science project developed with Machine Learning algorithms in order to predict the sales value of Rossmann pharmacies for the next 6 weeks. It was developed using the Python language, algorithms such as Random Forest and XGBoost and the practical visualization takes place through a Bot created in the Telegram application. The adopted model predicted a baseline scenario with consolidated gross revenue of €281.1M for the next 6 weeks. In the worst case, consolidated gross revenues will be €280.1M, and in the best case, €282M.
Project designed with the aim of helping a health insurance company rank the main potential customers to buy a new type of insurance for the company (car) and, therefore, carry out a Cross-Sell. To this end, Machine Learning and Learn to Rank techniques were used, so that it is possible to consult a list of customers most likely to purchase car insurance via Google Sheets, facilitating the company's communication strategy and optimizing its business. Finally, it was decided to make calls only to a part of the customer base with the highest purchasing propensity (46%) and the total profit was R$3.878.010,00, which represents a solution 1.44 times better than calling to 100% of the base.
Data science project developed with Machine Learning algorithms in order to predict the churn of customers of a banking company and develop strategies to avoid the phenomenon. It was developed using the Python language, algorithms such as Random Forest Classifier and XGBoost Classifier. The adopted model resulted in excellent performances in metrics such as the F1_score ( 85,1% ) and provided the formulation of an action plan to solve the churning problem based on sending discount coupons to customers according to their churn probability and the ROI (Return on Investment) maximization. At the end of the project, we achieve the final profit, as the average of the analyzed scenarios, of R$ 2.785.202,00, which represents an ROI of 361%.
In this project, a communications company named BuzzFeed is looking for ways to diagnose sarcasm in its news headlines, in order to avoid possible misunderstandings among readers in some sarcastic headlines. To achieve the objective, Natural Language Processing (NLP) and Machine Learning techniques were used, such as Word2Vec Embedding and Bidirectional LSTM Recurrent Neural Networks. The final model chosen presented excellent metrics, such as accuracy and AUC of 83%. Furthermore, the query to detect sarcastic titles can be done via Streamlit.
Predicting the price elasticity of e-commerce products and visualizing possible scenarios in Streamlit 📈:
The idea of the project is to study the concept of price elasticity and, therefore, the deep relationship between demand and product prices. In this sense, a Machine Learning model related to linear regression was developed, capable of predicting how much it is acceptable to increase/decrease the value of products, impacting demand, to try to find out if we could increase revenue. Visualization of possible scenarios after a discount or price increase can be viewed via Streamlit.
Project carried out with the focus on determining the group of customers most financially relevant for an e-commerce in accordance with the RFM model definitions. In this sense, several Clustering techniques in Machine Learning were used. The data products developed were a list with data from all customers and their clusters, a report with questions addressed by the marketing team and a Metabase dashboard fed with data updated through AWS services. The developed solution presents a group, made up of 7.2% of the customer base, which holds 40.20% of the company's total revenue, and an average revenue of $15,332.
Project with the objective of building a book recommendation system based on the behavior and preferences of each user with regard to purchase history and interactions with books. Recommendation and Machine Leaning techniques were used, such as NearestNeighbors and Cossine Similarity. The final model chosen was the cosine similarity technique, as it is simpler (does not require training) and presents high computational efficiency. Preview of recommended book names and images can be seen via Streamlit.
In this project, the concepts of python programming, data manipulation, strategic thinking and business logic, along with web development tools such as Streamlit and Github, were used to develop a management panel with the main metrics of a marketplace company food. The end result of the project was a panel hosted in a Cloud environment that assists the CEO in possible decision-making through insights generated from the analysis. The project is available through a link. The dashboard can be accessed from any device connected to the internet.
