Implementación de un modelo predictivo para la gestión de las ventas de OSM DE COLOMBIA SAS para el año 2025-1

Abstract

Inefficient sales forecasting is a recurring problem in many organizations, which employ manual and outdated methods to make predictions. This consumes valuable resources, both time and money, which can have serious economic consequences for companies. In this context, this thesis aims to develop a sales prediction model using machine learning techniques. This model will facilitate decision-making regarding import timing and costs, in addition to preventing lost sales.

The objective is to offer a solution that improves the accuracy of sales projections, considering that any margin of error could significantly affect financial results. In a highly competitive business environment, identifying patterns and relationships between indicators and sales is a key asset for a company's success. By implementing accurate and efficient sales forecasts, organizations not only recover losses but also drive sustainable growth, strengthening informed decision-making. The methodology used includes the selection of Machine Learning samples based on previously researched techniques, incorporating relevant variables from both the business environment and the external context. The sample obtained will be structured and processed in a Python IDLE environment, such as Google Colab.

After data normalization, which is a crucial step to ensure all variables are on a comparable scale and facilitate model training, a thorough statistical analysis will be carried out. This analysis allows for the identification of significant patterns and relationships in the data that may influence sales predictions and establishes the data structure so that the models can learn effectively.

Once the data is ready, several Machine Learning models, such as the Decision Tree, will be selected and trained. These models will be trained on one dataset and tested on another to assess their predictive capacity.The testing process will consist of comparing the model's predictions with actual sales figures or stock quotes. To measure the model's accuracy, metrics such as mean square error (MSE) and mean absolute error (MAE) will be used. Once the model is validated, the results will be presented through functional Python code, executed in tools such as Google Colab. The code will include instructions for running the model and visualizing the results, and a detailed guide will be provided to facilitate its use, even for users without experience in machine learning. This methodology ensures that the model is practical and applicable in real-world environments, helping organizations improve inventory management and sales forecasting using reliable data.