Forceasting hotel demands from medium post Data from Kaggle

Python version

import pandas as pd

df = pd.read_csv('../../../raw_data/hotel_bookings.csv')
df = df.rename(columns = {'arrival_date_year': 'year', 
                          'arrival_date_month': 'month', 
                          'arrival_date_day_of_month': 'day'})

# turn the months into numbers
def monthToNum(shortMonth):
    return {'January': 1, 'February': 2, 'March': 3,
            'April': 4, 'May': 5, 'June': 6,
            'July': 7, 'August': 8, 'September': 9, 
            'October': 10, 'November': 11, 'December': 12
    }[shortMonth]
    
df['month'] = df['month'].apply(monthToNum)
df['date'] = pd.to_datetime(df[['year', 'month', 'day']])

#filter hotel resort and remove cancelations
df = df[(df['is_canceled'] == 0) & (df['hotel'] == 'Resort Hotel')]
t_df = df.groupby(['date'])['hotel'].count().reset_index().rename(columns={'hotel':'y','date':'ds'})

import matplotlib.pyplot as plt

plt.plot(t_df.ds, t_df['y'])
plt.show()

In a very naive and limited way let’s create a training and testing set (keeping only last month for testing). This is obviously a poor choices as ideally we would need to test multiple months.

train_df = t_df.loc[(t_df['ds'] >= '2015-01-01') & (t_df['ds'] < '2017-08-01')]
test_df = t_df.loc[(t_df['ds'] >= '2017-08-01') & (t_df['ds'] > '2017-09-01')]

Using MAE (Mean Absolute Error) to estimate the suitability our model.

Hotel forecasting

Python version

Method 1. Arima.