In this homework, you will practice demand forecasting with both traditional

exponential smoothing, and ARIMA models, as well as the neural networks. You

will report the results for both MAE and MSE. Recall that exponential smooth and

ARIMA can only incorporate time series data but neural networks can

accommodate a wide range of features such as day of the week, month of the year,

temperature, etc.

Data Description: The data set has 6000 rows of data depicting the sales of the

products on a daily basis. The useful columns are: date, shop_id, item_id,

item_price, item_cnt_day.

The item_cnt_day gives us the demand of each item per day. If it is negative, it

means that a refund/replace has been initiated for that product on that day.

Divide the data into 80% training, 20% testing. You can use some part of the

training data for validation. Perform demand forecasting using exponential

smoothing, ARIMA and Neural Networks and report MAE and MSE for all the

models. Explain briefly about the comparison of the models based on the metrics

used.