credit : https://www.fleeteurope.com/fr/maas-safety-smart-mobility/news/london-bike-share-offers-corporate-solutions?a=FJA05&t%5B0%5D=Santander%20Cycles&curl=1

Introduction¶

The objective of this project is to perform multivariate time-series analysis to predict bike share demand.

Dataset Source¶

Kaggle - London bike sharing dataset

NB: This project also serves as my assignments for the course below -

Specialized Models: Time Series and Survival Analysis

Libraries & Configuration¶

In [ ]:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import tensorflow as tf

from pandas.plotting import register_matplotlib_converters
from sklearn.preprocessing import RobustScaler
from sklearn.metrics import mean_squared_error, r2_score
from keras.layers import CuDNNLSTM, Activation, Bidirectional, Dropout, Dense
from keras.models import Sequential 
from keras.callbacks import EarlyStopping

# configurations
%matplotlib inline
%config InlineBackend.figure_format='retina'
# add 2x retina display config
# https://github.com/ipython/ipython/pull/3381


# `pd.Timestamp` can be used in matplotlib plots without having to cast them to another type
# https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.plotting.register_matplotlib_converters.html
pd.plotting.register_matplotlib_converters()

sns.set(style='whitegrid', palette='muted', font_scale=1.5)
plt.rcParams['figure.figsize'] = 20, 9


RANDOM_SEED = 42

np.random.seed(RANDOM_SEED)
tf.random.set_seed(RANDOM_SEED)

Load Dataset¶

In [ ]:

!mkdir ~/.kaggle
!cp kaggle.json ~/.kaggle/
!chmod 600 ~/.kaggle/kaggle.json

# https://www.kaggle.com/hmavrodiev/london-bike-sharing-dataset
!kaggle datasets download -d hmavrodiev/london-bike-sharing-dataset
!unzip /content/london-bike-sharing-dataset.zip

Downloading london-bike-sharing-dataset.zip to /content
  0% 0.00/165k [00:00<?, ?B/s]
100% 165k/165k [00:00<00:00, 38.6MB/s]
Archive:  /content/london-bike-sharing-dataset.zip
  inflating: london_merged.csv

EDA¶

Metadata

timestamp - timestamp field for grouping the data
cnt - the count of a new bike shares
t1 - real temperature in C
t2 - temperature in C “feels like”
hum - humidity in percentage
wind_speed - wind speed in km/h
weather_code - category of the weather
is_holiday - boolean field - 1 holiday / 0 non holiday
is_weekend - boolean field - 1 if the day is weekend
season - category field meteorological seasons:
- 0-spring
- 1-summer
- 2-fall
- 3-winter
weather_code category description:
- 1 = Clear ; mostly clear but have some values with haze/fog/patches of fog/ fog in vicinity
- 2 = scattered clouds / few clouds
- 3 = Broken clouds
- 4 = Cloudy
- 7 = Rain/ light Rain shower/ Light rain
- 10 = rain with thunderstorm
- 26 = snowfall
- 94 = Freezing Fog

In [ ]:

df_pbs = pd.read_csv('/content/london_merged.csv', 
                     parse_dates=['timestamp'],
                     index_col='timestamp')
df_pbs

Out[ ]:

	cnt	t1	t2	hum	wind_speed	weather_code	is_holiday	is_weekend	season
timestamp
2015-01-04 00:00:00	182	3.0	2.0	93.0	6.0	3.0	0.0	1.0	3.0
2015-01-04 01:00:00	138	3.0	2.5	93.0	5.0	1.0	0.0	1.0	3.0
2015-01-04 02:00:00	134	2.5	2.5	96.5	0.0	1.0	0.0	1.0	3.0
2015-01-04 03:00:00	72	2.0	2.0	100.0	0.0	1.0	0.0	1.0	3.0
2015-01-04 04:00:00	47	2.0	0.0	93.0	6.5	1.0	0.0	1.0	3.0
...	...	...	...	...	...	...	...	...	...
2017-01-03 19:00:00	1042	5.0	1.0	81.0	19.0	3.0	0.0	0.0	3.0
2017-01-03 20:00:00	541	5.0	1.0	81.0	21.0	4.0	0.0	0.0	3.0
2017-01-03 21:00:00	337	5.5	1.5	78.5	24.0	4.0	0.0	0.0	3.0
2017-01-03 22:00:00	224	5.5	1.5	76.0	23.0	4.0	0.0	0.0	3.0
2017-01-03 23:00:00	139	5.0	1.0	76.0	22.0	2.0	0.0	0.0	3.0

17414 rows × 9 columns

In [ ]:

df_pbs.info()

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 17414 entries, 2015-01-04 00:00:00 to 2017-01-03 23:00:00
Data columns (total 9 columns):
 #   Column        Non-Null Count  Dtype  
---  ------        --------------  -----  
 0   cnt           17414 non-null  int64  
 1   t1            17414 non-null  float64
 2   t2            17414 non-null  float64
 3   hum           17414 non-null  float64
 4   wind_speed    17414 non-null  float64
 5   weather_code  17414 non-null  float64
 6   is_holiday    17414 non-null  float64
 7   is_weekend    17414 non-null  float64
 8   season        17414 non-null  float64
dtypes: float64(8), int64(1)
memory usage: 1.3 MB

In [ ]:

df_pbs.describe().T

Out[ ]:

	count	mean	std	min	25%	50%	75%	max
cnt	17414.0	1143.101642	1085.108068	0.0	257.0	844.0	1671.75	7860.0
t1	17414.0	12.468091	5.571818	-1.5	8.0	12.5	16.00	34.0
t2	17414.0	11.520836	6.615145	-6.0	6.0	12.5	16.00	34.0
hum	17414.0	72.324954	14.313186	20.5	63.0	74.5	83.00	100.0
wind_speed	17414.0	15.913063	7.894570	0.0	10.0	15.0	20.50	56.5
weather_code	17414.0	2.722752	2.341163	1.0	1.0	2.0	3.00	26.0
is_holiday	17414.0	0.022051	0.146854	0.0	0.0	0.0	0.00	1.0
is_weekend	17414.0	0.285403	0.451619	0.0	0.0	0.0	1.00	1.0
season	17414.0	1.492075	1.118911	0.0	0.0	1.0	2.00	3.0

Missing Values¶

In [ ]:

pd.date_range(df_pbs.index.min(), df_pbs.index.max()).difference(df_pbs.index)

Out[ ]:

DatetimeIndex(['2015-09-12', '2015-11-18', '2016-02-08', '2016-06-24',
               '2016-07-31', '2016-09-02', '2016-09-03'],
              dtype='datetime64[ns]', freq=None)

Since the ratio of missing dates is very low (~0.1%), no action is taken to fill missing values

Feature Extraction¶

In [ ]:

# add hour of the day 
df_pbs['hour'] = df_pbs.index.hour

# add day of month
df_pbs['day_of_month'] = df_pbs.index.day

# add day of week
df_pbs['day_of_week'] = df_pbs.index.dayofweek

# add month
df_pbs['month'] = df_pbs.index.month

In [ ]:

sns.lineplot(x=df_pbs.index, y="cnt", data=df_pbs);

The data is crowded. The grpah would be clear if it's plotted on monthly basis

Monthly Basis¶

Seasonal Changes (2015-01-31 to 2017-01-31)¶

In [ ]:

df_pbs.resample('M').sum()

Out[ ]:

	cnt	t1	t2	hum	wind_speed	weather_code	is_holiday	is_weekend	season
timestamp
2015-01-31	546639	4101.000000	2255.500000	51005.00	12502.000000	2211.0	0.0	192.0	2013.0
2015-02-28	543594	3721.500000	1868.000000	51950.00	9873.500000	2409.0	0.0	192.0	2013.0
2015-03-31	695934	6093.500000	4558.500000	51832.25	12938.250000	1940.0	0.0	215.0	0.0
2015-04-30	831582	7873.000000	7112.500000	48468.00	10988.500000	1642.0	48.0	191.0	0.0
2015-05-31	895413	10091.500000	9803.500000	47939.00	14268.500000	1900.0	48.0	240.0	0.0
2015-06-30	1033252	12151.500000	12073.000000	44371.50	12398.000000	1457.0	0.0	192.0	717.0
2015-07-31	1120687	14107.000000	14060.000000	46632.50	13067.000000	1710.0	0.0	192.0	740.0
2015-08-31	1033951	13830.500000	13816.500000	51913.50	10326.000000	1948.0	24.0	240.0	744.0
2015-09-30	892478	10371.000000	10357.500000	52002.00	9394.000000	1683.0	0.0	185.0	1422.0
2015-10-31	865046	9474.000000	9321.500000	57887.00	8671.166667	2078.0	0.0	215.0	1472.0
2015-11-30	677332	8259.500000	7605.000000	56344.50	14191.000000	2418.0	0.0	216.0	1422.0
2015-12-31	602838	9000.500000	8660.000000	56448.50	15452.500000	2367.0	48.0	188.0	2220.0
2016-01-31	582518	5459.500000	3872.000000	58230.50	12996.500000	2424.0	24.0	240.0	2232.0
2016-02-29	592910	4695.500000	3033.000000	49847.00	13058.500000	2054.0	0.0	190.0	2064.0
2016-03-31	656527	5344.000000	3684.500000	53722.50	11778.000000	2177.0	48.0	177.0	0.0
2016-04-30	768845	6804.500000	5708.500000	48643.50	11067.000000	1901.0	0.0	216.0	0.0
2016-05-31	1001935	10663.000000	10537.500000	50815.50	11001.500000	1852.0	48.0	216.0	0.0
2016-06-30	933878	11602.500000	11578.500000	53668.50	9745.000000	2204.0	0.0	192.0	705.0
2016-07-31	1165527	14467.000000	14453.500000	48719.50	11847.000000	1619.0	0.0	239.0	741.0
2016-08-31	1137314	14699.000000	14686.500000	47944.00	11860.500000	1511.0	24.0	192.0	740.0
2016-09-30	998619	12813.000000	12800.000000	48022.00	9580.000000	1622.0	0.0	180.0	1366.0
2016-10-31	935706	9363.333333	9239.833333	56942.00	9328.666667	1919.0	0.0	239.0	1486.0
2016-11-30	703859	5827.000000	4499.500000	56243.00	10499.000000	2128.0	0.0	191.0	1438.0
2016-12-31	651908	5967.500000	4925.000000	63803.50	9216.500000	2005.0	48.0	216.0	2232.0
2017-01-31	37680	339.000000	114.000000	6071.50	1061.500000	235.0	24.0	24.0	216.0

In [ ]:

df_pbs.resample('M').sum().plot(y='cnt');
plt.xlabel('Month')
plt.ylabel('Bike Share Count (million(s))')
plt.suptitle('Bike Share Monthly Basis Trend')
plt.show();

# df_pbs_month = df_pbs.resample('M').sum()
# sns.lineplot(x=df_pbs_month.index, y='cnt', data=df_pbs_month);

From the graph above it is clear that it has a strong seasonality component. Summer months are good for bike share business

Autocorrelation¶

We can quantify the strength and type of relationship between observations and their lags. In statistics, this is called correlation, and when calculated against lag values in time series, it is called autocorrelation (self-correlation)

A correlation value calculated between two groups of numbers, such as observations and their lag1 values, results in a number between -1 and 1. The sign of this number indicates a negative or positive correlation respectively. A value close to zero suggests a weak correlation, whereas a value closer to -1 or 1 indicates a strong correlation

In [ ]:

pd.plotting.autocorrelation_plot(df_pbs.cnt);

Introduction¶

Dataset Source¶

Libraries & Configuration¶

Load Dataset¶

EDA¶

Missing Values¶

Feature Extraction¶

Bike Share Over Time¶

Monthly Basis¶

Seasonal Changes (2015-01-31 to 2017-01-31)¶

Autocorrelation¶