Singtel High Accuracy Forecast Web Application

Our project was developed to solve the problems presented by the first problem statement by Singtel. It includes all the features necessary to fulfill all the objectives.

To achieve a high forecast accuracy, we first prepare the data. We include several features, such as a lag-1 feature, which is the previous actual sales amount, and a weekly feature, which is the number of weeks it has been released, and also calculate the total amount of offers given within the week and use it as a feature. Finally, we feed the model to an XGBoost Regressor. Since the number of data is small, we train by concatenating all the devices in only one model. After fine-tuning it, we conclude that the best hyperparameters for us are n_estimator=1000, max_depth=6, subsample=0.7, and colsample_bytree=0.9.

To maintain the AI model and to make the app scalable, we stored the data needed for the prediction model to make decisions in a database. Selected users are able to write onto the database, changing data, which will result in different and more accurate predictions as the number of data available increases. All stakeholders are able to enjoy the collected and predicted data in formatted graphs and tables.

The front end of this application is made using a React framework called Next JS and the back end uses a popular Python framework called Django Rest Framework.

We included an inventory page to allow the company to store information on individual phones, as well as their amounts in the company’s inventory. The number of stocks would be updated automatically following phone transactions.

In order to facilitate the usage of the dashboard by various stakeholders, we will implement a permissions system that grants access and functionality based on the user's role within the company. This will ensure that only authorized personnel can view and modify information related to sales and inventory.

Ultimately, our dashboard also focuses on a user-friendly display of the website, to ensure that users have a pleasant experience using it.

Front-End Repo: Front-end Repo

Back-End Repo: Back-end Repo