title	author	Update	output
HOW TO USE TENSORFLOW 2.X TO PREDICT BIKE RENTAL USING ANNs (Regression)	Dr LEBEDE Ngartera	1--07--2024	html_document pdf_document toc number_sections true true toc true

HOW TO USE TENSORFLOW 2.X TO PREDICT BIKE RENTAL USING ANNs (Regression)

0) INTRODUCTION: PROJECT STATEMENT

In pursuit of enhancing bike rental management, this project focuses on predicting rental usage by leveraging key environmental factors such as temperature, humidity, and wind speed. By harnessing the power of machine learning, the objective is to construct a multi-layer perception network. This sophisticated neural network architecture will enable the system to analyze intricate patterns within the input variables, providing a comprehensive understanding of their impact on bike rental trends. The ultimate goal is to develop a predictive model that not only captures the nuances of weather-related influences but also contributes to more efficient and informed decision-making in the bike rental domain. Through the implementation of advanced algorithms, this initiative aims to optimize resource allocation and enhance user experience in the dynamic realm of bike-sharing services.

Data Reference:

1. SuperDataScience
2. Laboratory of Artificial Intelligence and Decision Support (LIAAD), University of Porto INESC Porto, Campus da FEUP Rua Dr. Roberto Frias, 378 4200 - 465 Porto, Portugal

Data Description:

1. instant: record index
2. dteday : date
3. season : season (1:springer, 2:summer, 3:fall, 4:winter)
4. yr : year (0: 2011, 1:2012)
5. mnth : month ( 1 to 12)
6. hr : hour (0 to 23)
7. holiday : wether day is holiday or not (extracted from http://dchr.dc.gov/page/holiday-schedule)
8. weekday : day of the week
9. workingday : if day is neither weekend nor holiday is 1, otherwise is 0. weathersit :

a) Clear, Few clouds, Partly cloudy

b) Mist + Cloudy, Mist + Broken clouds, Mist + Few clouds, Mist

c) Light Snow, Light Rain + Thunderstorm + Scattered clouds, Light Rain + Scattered clouds

d) Heavy Rain + Ice Pallets + Thunderstorm + Mist, Snow + Fog

1. temp : Normalized temperature in Celsius. The values are divided to 41 (max)
2. hum: Normalized humidity. The values are divided to 100 (max)
3. windspeed: Normalized wind speed. The values are divided to 67 (max)
4. casual: count of casual users
5. registered: count of registered users
6. cnt: count of total rental bikes including both casual and registered

I) IMPORT LIBRARIES

II) IMPORT DATASETS

We will need to mount your drive using the following commands:

For more information regarding mounting, please check this out: https://stackoverflow.com/questions/46986398/import-data-into-google-colaboratory

from google.colab import drive drive.mount('/content/drive')

Mounted at /content/drive We have to include the full link to the csv file containing your dataset

bike = pd.read_csv('/content/drive/My Drive/bike_sharing_daily.csv')

bike

bike.info()

III) CLEAN UP DATASET

sns.heatmap(bike.isnull())

bike = bike.drop(labels = ['instant'], axis = 1)

bike.index = pd.DatetimeIndex(bike.dteday)

bike

IV) VISUALIZE DATASET

bike['cnt'].asfreq('W').plot(linewidth = 3)

plt.title('Bike Usage Per week')

plt.xlabel('Week')

plt.ylabel('Bike Rental')

bike['cnt'].asfreq('M').plot(linewidth = 3)

plt.title('Bike Usage Per Month')

plt.xlabel('Month')

plt.ylabel('Bike Rental')

bike['cnt'].asfreq('Q').plot(linewidth = 3)

plt.title('Bike Usage Per Quarter')

plt.xlabel('Quarter')

plt.ylabel('Bike Rental')

sns.pairplot(bike)

X_numerical = bike[['temp', 'hum', 'windspeed', 'cnt']]

X_numerical

sns.pairplot(X_numerical)

sns.heatmap(X_numerical.corr(), annot =True)

V) CREATE TRAINING AND TESTING DATASET

X_cat = bike[['season', 'yr', 'mnth', 'holiday', 'weekday', 'workingday', 'weathersit']]

X_cat

from sklearn.preprocessing import OneHotEncoder onehotencoder = OneHotEncoder() X_cat = onehotencoder.fit_transform(X_cat).toarray()

X_cat.shape

(731, 32)

X_cat = pd.DataFrame(X_cat)

X_numerical

X_numerical = X_numerical.reset_index()

X_all = pd.concat([X_cat, X_numerical], axis = 1)

X_all account_circle

X_all = X_all.drop('dteday', axis = 1)

X_all

X = X_all.iloc[:, :-1].values y = X_all.iloc[:, -1:].values [ ] 1 X.shape

(731, 35)

y.shape

(731, 1)

from sklearn.preprocessing import MinMaxScaler scaler = MinMaxScaler() y = scaler.fit_transform(y)

from sklearn.model_selection import train_test_split X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2)

X_train.shape

(584, 35)

X_test.shape

(147, 35)

VI) TRAIN THE MODEL

model = tf.keras.models.Sequential()

model.add(tf.keras.layers.Dense(units=100, activation='relu', input_shape=(35, )))

model.add(tf.keras.layers.Dense(units=100, activation='relu'))

model.add(tf.keras.layers.Dense(units=100, activation='relu'))

model.add(tf.keras.layers.Dense(units=1, activation='linear'))

model.summary()

Model: "sequential"

---

Layer (type) Output Shape Param

dense (Dense) (None, 100) 3600

dense_1 (Dense) (None, 100) 10100

dense_2 (Dense) (None, 100) 10100

dense_3 (Dense) (None, 1) 101

=================================================================

Total params: 23,901

Trainable params: 23,901

Non-trainable params: 0

---

model.compile(optimizer='Adam', loss='mean_squared_error')

epochs_hist = model.fit(X_train, y_train, epochs = 20, batch_size = 50, validation_split = 0.2)

Epoch 1/20 10/10 [==============================] - 1s 24ms/step - loss: 0.1408 - val_loss: 0.0675 Epoch 2/20 10/10 [==============================] - 0s 5ms/step - loss: 0.0410 - val_loss: 0.0331 Epoch 3/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0199 - val_loss: 0.0213 Epoch 4/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0135 - val_loss: 0.0176 Epoch 5/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0104 - val_loss: 0.0182 Epoch 6/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0095 - val_loss: 0.0169 Epoch 7/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0076 - val_loss: 0.0156 Epoch 8/20 10/10 [==============================] - 0s 6ms/step - loss: 0.0064 - val_loss: 0.0154 Epoch 9/20 10/10 [==============================] - 0s 8ms/step - loss: 0.0057 - val_loss: 0.0159 Epoch 10/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0053 - val_loss: 0.0148 Epoch 11/20 10/10 [==============================] - 0s 6ms/step - loss: 0.0049 - val_loss: 0.0144 Epoch 12/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0047 - val_loss: 0.0143 Epoch 13/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0043 - val_loss: 0.0143 Epoch 14/20 10/10 [==============================] - 0s 5ms/step - loss: 0.0042 - val_loss: 0.0147 Epoch 15/20 10/10 [==============================] - 0s 5ms/step - loss: 0.0042 - val_loss: 0.0149 Epoch 16/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0041 - val_loss: 0.0152 Epoch 17/20 10/10 [==============================] - 0s 5ms/step - loss: 0.0038 - val_loss: 0.0145 Epoch 18/20 10/10 [==============================] - 0s 5ms/step - loss: 0.0035 - val_loss: 0.0159 Epoch 19/20 10/10 [==============================] - 0s 5ms/step - loss: 0.0035 - val_loss: 0.0153 Epoch 20/20 10/10 [==============================] - 0s 7ms/step - loss: 0.0034 - val_loss: 0.0145

VII) EVALUATE THE MODEL

epochs_hist.history.keys()

dict_keys(['loss', 'val_loss'])

plt.plot(epochs_hist.history['loss'])

plt.plot(epochs_hist.history['val_loss'])

plt.title('Model Loss Progress During Training')

plt.xlabel('Epoch')

plt.ylabel('Training and Validation Loss')

plt.legend(['Training Loss', 'Validation Loss'])

y_predict = model.predict(X_test) plt.plot(y_test, y_predict, "^", color = 'g') plt.xlabel('Model Predictions') plt.ylabel('True Values')

y_predict_orig = scaler.inverse_transform(y_predict) y_test_orig = scaler.inverse_transform(y_test)

plt.plot(y_test_orig, y_predict_orig, "^", color = 'b') plt.xlabel('Model Predictions') plt.ylabel('True Values')

k = X_test.shape[1] n = len(X_test) n

147

from sklearn.metrics import r2_score, mean_squared_error, mean_absolute_error

from math import sqrt

RMSE = float(format(np.sqrt(mean_squared_error(y_test_orig, y_predict_orig)),'.3f'))

MSE = mean_squared_error(y_test_orig, y_predict_orig)=

MAE = mean_absolute_error(y_test_orig, y_predict_orig)

r2 = r2_score(y_test_orig, y_predict_orig)

adj_r2 = 1-(1-r2)*(n-1)/(n-k-1)

RMSE = 855.646

MSE = 732130.1093405469

MAE = 614.8364324245323

R2 = 0.805684986705684

Adjusted R2 = 0.744414487018287

Author

Lebede Ngartera