Fare & Duration Prediction for NYC Yellow Cabs

• This Project is for course IST 718: Big Data Analytics @ Syracuse University
• Professor: Daniel Acuna deacuna@syr.edu

Objective and Focus:

• Prediction of Fare and Estimated duration of a ride from Point A to B
• Does weather and holidays really affect the profitability of driver?

Philosophy:

• Information Unavailability for drivers
• Fare negotiation:The yellow cab drivers spend time in making a proposal and waiting to make a deal post the effort of finding a customer

Requirements:

• Python 3.7.0.
• Pandas
• Numpy
• Pyspark 2.4.0
• Spark 2.4.0
• xgboost
• sklearn
• matplotlib.pyplot
• seaborn
• RandomForestRegressor
• GBTRegressor
• RegressionEvaluator
• VectorAssembler
• ascii_letters
• datetime
• StandardScaler
• MinMaxScaler
• Vector

Installation:

pip install pyspark

Project Code : In addition to the requirements, we need to Change these lines to read files locally :

Data Description:

• Full Merged Dataset : NYC Yellow Taxi Trip Records + USA Federal Holidays Dataset + NYC Weather Data
• Number of data points : 12 M
• Training data points out of Sample Dataset: 600 K
• Total Features: After cleaning, the total features are 61
• Cleanliness: Merging, Date time conversion, Imputation, Dummy variables, Standardization
• Label/output to predict: Duration and Total Amount

Findings

• The average duration of a trip is about 15-18 minutes • The average total amount is between $10-$19
• The average fare price is highest for the month of October
• Longest rides occur at 3:00 PM in the evenings and at 6:00 AM in the morning, the distance of the rides is the shortest
• Overall, our models for predicting taxi pickups’ total amount and duration in New York city performed well
• The gradient boosting regression model closely followed by the random forest model performed the best
• This was likely due to their unique ability to capture complex feature dependencies
• The rmse for gradient boosting regression model was achieved to be ~$3.71178 for amount and ~87.3163 sec for total duration
• Our results and error analysis for the most part supported our intuitions for the usefulness of features with the exception of holiday feature which was not found important for model performance
• Our model can be used by city planners and yellow taxi drivers in determining where to position yellow taxi cabs and understanding patterns in ridership
• In our future work we plan to implement two models: neural network regression for its capability to automatically tune and model feature interactions Although we used k-means clustering, we aim to achieve better results in the future

ProjectCode

Project Code

Change these lines :

fulldata = pd.read_csv("/Users//IST718 Dropbox/IST 718 Project/Taxi Data/2015-01_100k.csv")
weatherdata = pd.read_excel("/Users//IST718 Dropbox/IST 718 Project/Weather Data/2015_weather.xlsx")
holidaysdata = pd.read_excel("/Users//IST718 Dropbox/IST 718 Project/Holiday Data/holidays.xlsx")

into these lines :

fulldata = pd.read_csv(os.environ.get("TAXI_DATA"))
weatherdata = pd.read_excel(os.environ.get("WEATHER_DATA"))
holidaysdata = pd.read_excel(os.environ.get("HOLIDAY_DATA"))

### Setup the environment variables for Data Path

Append these lines into this file `~/.bash_profile` 
CAUTION: Do not overwrite bash profile, just add these lines to it !

The contents of this file should look like this :

```bash
export TAXI_DATA="/Users//IST718 Dropbox/IST 718 Project/Taxi Data/2015-01_100k.csv"
export WEATHER_DATA="/Users//IST718 Dropbox/IST 718 Project/Weather Data/2015_weather.xlsx"
export HOLIDAY_DATA="/Users//IST718 Dropbox/IST 718 Project/Holiday Data/holidays.xlsx"

How to launch the anaconda navigator

From a new command line terminal, run anaconda-navigator everytime.

$ anaconda-navigator

Instructions for Github

Please push changes ONLY to your respective notebooks to avoid github conflicts in ipynb files

How to compare two notebooks

install xcode-select like this:

$ xcode-select --install

and then, install nbdiff

and then, run like following from command line :

MacBook-Pro:ProjectCode apsharma$ nbdiff IST\ 718_Apurva.ipynb IST\ 718_Shivanshi.ipynb 
nbdiff IST 718_Apurva.ipynb IST 718_Shivanshi.ipynb

--- IST 718_Apurva.ipynb  2018-10-28 10:16:25.199350
+++ IST 718_Shivanshi.ipynb  2018-10-28 10:21:35.103231
## deleted /cells/2:
-  code cell:
-    execution_count: 7
-    metadata (known keys):
-      collapsed: True
-    source:
-      #import pandas as pd
-      #fulldata=pd.read_csv("C:/Users/SHIVANSHI/IST718 Dropbox/IST 718 Project/Taxi Data/2015-01_100k.csv")
-      #weatherdata = pd.read_excel("C:/Users/SHIVANSHI/IST718 Dropbox/IST 718 Project/Weather Data/2015_weather.xlsx")
-      #holidaysdata = pd.read_excel("C:/Users/SHIVANSHI/IST718 Dropbox/IST 718 Project/Holiday Data/holidays.xlsx")

## modified /cells/3/source:
@@ -1,4 +1,4 @@
 #import pandas as pd
-#fulldata=pd.read_csv("C:/Users/SHAMA/IST718 Dropbox/IST 718 Project/Taxi Data/2015-01_100k.csv")
-#weatherdata = pd.read_excel("C:/Users/SHAMA/IST718 Dropbox/IST 718 Project/Weather Data/2015_weather.xlsx")
-#holidaysdata = pd.read_excel("C:/Users/SHAMA/IST718 Dropbox/IST 718 Project/Holiday Data/holidays.xlsx")
+#fulldata=pd.read_csv("C:/Users//IST718 Dropbox/IST 718 Project/Taxi Data/2015-01_100k.csv")
+#weatherdata = pd.read_excel("C:/Users//IST718 Dropbox/IST 718 Project/Weather Data/2015_weather.xlsx")
+#holidaysdata = pd.read_excel("C:/Users//IST718 Dropbox/IST 718 Project/Holiday Data/holidays.xlsx")

and to view these changes in browser instead , run like following :

$ nbdiff-web IST\ 718_Apurva.ipynb IST\ 718_Shivanshi.ipynb 

For more on nbdiff and merging conflicts in ipython notebooks, refer: nbdiff documentation