Sales forecasting

Proper sales forecasting means better budget management. However, sales forecasting is often based on basic statistics such as oridinary least square regression, not often suited for this type of data. We deliver deep learning models to accurately forecast your future sales.
Below is an example of sales forecasting we do.

1. INITIAL QUESTION

A bike rental company observed that they rent much more bikes than the last year at the same period. They want to know if this tendency will persist, in which case they will have to invest 100K€ for the purchasing of new bikes, in order to respond to this increasing demand.

2. DATA COLLECTION

The amount of rented bikes for a given date is supplied by the company. For each date, we add meta-data gathered online such as the season, whether it is a working day or not, the weather (temperature, air moisture, wind speed, etc.).

3. DATA ANALYSIS

We test several algorithms and choose an artificial neural network. We optimize the network structure using hyper-parameter tuning methods. Part of the dataset was used to validate the model and another part was used to test the model, in order to avoid overfitting. Overfitting is the property of a model to perform better on the training set than on the test set, meaning that the model cannot be used in production.

  

4. RESULTS

Our model successfully predicted a decrease in the amount of rented bikes, even though the test set exhibits patterns the model was not trained to recognize. This behaviour is a property of deep learning algorithms, that are able to “understand” complex interactions between parameters.

  

Sales forecasting Machine learning DataTailors

  

5. CONCLUSION

Our model is powerful and can predict patterns it wasn’t trained to recognize beforehand. We advise the company not to invest immediately in new bikes. We propose the company an easy-to-use rent forecasting software they can use autonomously.