This article aims to predict the count of casual users (feature casual), count of registered users (feature registered), and the total count of both causal and registered users (feature cnt) using the multi-regression model.
You can go to this webpage to find the source data day.csv and also you can find Bike Sharing Dataset Analytics -- Daily Data.py here.

Exploring the data

import pandas as pd
day = pd.read_csv("day.csv")
day.head()

Visualization

import matplotlib.pyplot as plt
import seaborn as sns

result = day[["mnth", "casual", "registered", "cnt"]].groupby(["mnth"]).mean()
result = (
    result.stack()
    .reset_index()
    .set_index("mnth")
    .rename(columns={"level_1": "cat", 0: "people per day"})
)
cat    = ["season", "yr", "holiday", "weekday", "workingday", "weathersit"]

sns.set(style="ticks", palette="pastel")
f, axes = plt.subplots(3, 3, sharey=False, figsize=(15, 12))
ax      = plt.subplot2grid((3, 3), (0, 0), colspan=3)
sns.barplot(x=result.index, y="people per day", data=result, hue="cat", ax=ax)
for ind, val in enumerate(cat):
    result = round(day[[val, "casual", "registered", "cnt"]].groupby([val]).mean())
    result = (
        result.stack()
        .reset_index()
        .set_index(val)
        .rename(columns={"level_1": "cat", 0: "people per day"})
    )
    sns.barplot(
        x    = result.index,
        y    = "people per day",
        data = result,
        hue  = "cat",
        ax   = axes[ind // 3 + 1, ind % 3],
    )
f.tight_layout(pad=3.0)
plt.show()

result = day[["casual", "registered", "cnt"]].set_index(day["mnth"])
result = (
    result.stack()
    .reset_index()
    .set_index("mnth")
    .rename(columns={"level_1": "cat", 0: "people"})
)

f, axes = plt.subplots(3, 3, sharey=False, figsize=(20, 12))
ax      = plt.subplot2grid((3, 3), (0, 0), colspan=3)
sns.violinplot(x=result.index, y="people", hue="cat", data=result, cut=0, ax=ax)
for ind, val in enumerate(cat):
    result = day[["casual", "registered", "cnt"]].set_index(day[val])
    result = (
        result.stack()
        .reset_index()
        .set_index(val)
        .rename(columns={"level_1": "cat", 0: "people"})
    )
    sns.violinplot(
        x    = result.index,
        y    = "people",
        hue  = "cat",
        data = result,
        cut  = 0,
        ax   = axes[ind // 3 + 1, ind % 3],
    )
f.tight_layout(pad=3.0)
plt.show()

The number of people who rental a bike every year is increasing, which indicates that riding bicycle is getting popular. This trend could be got if we fit a model on date and year. But here, since we don't have data of enough years, so we maily focus on the other features in our analysis.

Linear Regression

Split Data

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

day[cat] = day[cat].apply(lambda x: x.astype("category"))
dummies  = pd.get_dummies(day[cat], drop_first=True)
print(dummies.shape)

conti_predictors = ["temp", "atemp", "hum", "windspeed"]
print(day[conti_predictors].shape)
X = pd.concat([day[conti_predictors], dummies], axis=1)
y = day.iloc[:, 13:]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25)
sc                               = StandardScaler()
X_train[conti_predictors]        = sc.fit_transform(X_train[conti_predictors])
X_test [conti_predictors]        = sc.transform(X_test[conti_predictors])

(731, 24)
(731, 4)

Train Model

from sklearn.linear_model import LinearRegression
from dmba import adjusted_r2_score

lr     = LinearRegression()
lr.fit(X_train, y_train)
y_pred = lr.predict(X_test)
y_pred = pd.DataFrame(y_pred, columns=["casual", "registered", "cnt"]).astype(int)
print(y_pred)
print(adjusted_r2_score(y_test, y_pred, lr))

0.5664326431246609

for i in range(y_pred.shape[0]):
    if y_pred["casual"].iloc[i] < 0:
        y_pred["casual"].iloc[i] = 0
    if y_pred["registered"].iloc[i] < 0:
        y_pred["registered"].iloc[i] = 0
        y_pred["cnt"].iloc[i]        = y_pred["casual"].iloc[i] + y_pred["registered"].iloc[i]
print(y_pred)
print(adjusted_r2_score(y_test, y_pred, lr))

0.5761640858277055