fourierers

Inspiration

Learning about IA and machine learning.

What it does

It predicts the future of the target variable by training a model through a dataset. Secondly and most important it it explains and shows clearly the importance of each variable.

How we built it

From the dataset, we were able to create a classification model that attempts to predict future sales based on the variables in the dataset. First, we split the dataset, taking 80% to train the model and 20% to test it and check the validity of the results. The machine learning model we use is a Random Forest, which works by combining the output of multiple decision trees to arrive at a single result. Initially, we created our model without any filters and looked at correlations, whether there were majority classes, among other things. We noticed that there was a small imbalance between won and lost sales, and to optimize it, we eliminated this imbalance. Additionally, we used GridSearch to systematize the combinations and make it more robust. We started with an f1_score of 0.71, and after applying filters and using GridSearch, it was 0.84.

Now we compile the results in a table where we present the metrics of the optimized model. Precision is the number of times the model predicts the actual outcome correctly, that is, when the result is 0, it is correct 86% of the time. And speaking of 1’s, it is correct 82% of the time. Recall indicates which class the model identifies better; in our case, the model identifies 1’s better than 0’s, even though it makes more mistakes with them. The F1-score corresponds to a measure that combines precision and recall into a single metric. The model is slightly better with 0’s, although for class 1 it has a higher recall.

Challenges we ran into

Learning about explainability techniques was more difficult than we thought since nobody aroun us seemed to be familiar with the concept

Accomplishments that we're proud of

We accomplished a pretty good accuracy in our machine learning model by using new techniques that we learned throughout this weekend

What we learned

We learned how to properly train a model

What's next for fourierers

We would like to keep learning about data scince for next datathons since we are not very familiar with the concepts and it is a good way to get in touch with a new and fascinating engineering world.