The exciting challenge of the 6th edition of the Novartis España Datathon. Tasked with unraveling the complexities of pharmaceutical sales data spread across various countries, brands, and years (2013 to 2021), we embarked on a journey through a vast ocean of data.
Our mission? To forecast the distribution of sales throughout the entire year of 2022. This quest in Time Series analysis led us to meticulously dissect data trends, seasonality, and cyclical patterns, alongside rigorous preprocessing steps.
Navigating the time-dependent intricacies of the data, we engineered a suite of features including lag components and rolling statistical estimators. Our strategy? Harnessing the power of the LightGBM Model. This choice was driven by its remarkable efficiency – delivering robust performance with minimal memory usage and swift training times.
Though we didn't advance to the final round, the performance of our model amidst the fiercely competitive environment was a win in itself. A notable mention goes to the newly launched Latitude software, a pivotal tool in our arsenal that brought clarity and vivid visuals to our data exploration phase.
Reflecting on the event, it was more than just a competition; it was a rich learning journey. We delved deep into the nuances of Time Series data pipelines, emerging with newfound knowledge and skills. And, as always, the thrill of a Datathon was amplified by the camaraderie and collective spirit of my teammates.
In sum, the Novartis Datathon was not just an event; it was an opportunity to grow, learn, and enjoy every moment of the challenge.