Machine-LearningAutomatidata

Machine-Learning Automatidata

We are creating a machine learning model as a consultant in a fiction consulting firm, ‘Automatida’, for the ‘New York City Taxi and Limousine Commission’ to ancipate whether or not a customer will be a generous tipper to their drivers. We will implement feature engineering, model development, and evaluation to figure out what factors affect that.

Background on the Automatidata scenario

Automatidata works with its clients to transform their unused and stored data into useful solutions, such as performance dashboards, customer-facing tools, strategic business insights, and more. They specialize in identifying a client’s business needs and utilizing their data to meet those business needs.

Automatidata is consulting for the New York City Taxi and Limousine Commission (TLC). New York City TLC is an agency responsible for licensing and regulating New York City’s taxi cabs and for-hire vehicles. The agency has partnered with Automatidata to develop a regression model that helps estimate taxi fares before the ride, as well as tipping profiles based on data that TLC has gathered. The purpose is for TLC to better create algorithms that benefits both the quality of the customer’s ride experience, as well as Taxi driver’s work experience.

The TLC data comes from over 200,000 taxi and limousine licensees, making approximately one million combined trips per day.

Note: This project’s dataset was created for pedagogical purposes and may not be indicative of New York City taxi cab riders’ behavior.

Assignment

We will create a machine learning model for the TLC data to predict whether or not a customer is a generous tipper, what features affect that, and calculate the machine learning model’s accuracy.

In this project, we will complete the following deliverables:

Data Clean and Organize

Design and implement a machine learning model

Analyze important factors

Draft an executive summary of your results

Data Cleaning and Organizing

We managed to merge 2 dataframes, delete unnecessary columns, and create new neccessary columns from main columns for the research. This coding process is shownn in the Data Cleaning Code page. This resulted in us having a column,’generous’, which depicted whether a customer was considered to be generous by the percentage of tip they gave compared to the total cost of the trip. This would be very useful in the proceeding of the research.

Design and implement a machine learning model

We created a Random Forest Classifier and an XGB Classifier and compare the F1 Score to identity the most accurate model in predicting the kind of customers that would be the most generous tippers.

We first began by testing the balance of the data, comparing the amount of data that had generous customers and the amount that didn’t. This was to make sure our model would have enough data of generous and ungenerous tippers to be able to accurately understand and predict what factors makes those profile of tippers without reporting false or unrepresentative predictions. Our outcome fortunately came out to generous tippers containing 52% of the data, as seen in the image below, which is about even; therefore, there would be no misinterpretation when creating the model.

Knowing the data wasn’t biased, we decided to build a Random Forest Classifier and an XGB Classifier. We also fitted training set on both Classifiers for better results.

Seeing that our F1 scores in both classifiers are over .50, this means using either models gives us a better chance of predicting who will be a generous tipper than, say a coin flip. However, given that our Rainforest Classifier has a higher F1 score than our XGB model, we opted for the Rain Forest Classifier as a higher F1 score means this model is better for predicting than the XGB model.

We then created a Confusion Matrix to understand the characteristics of the Rain Forest Classifyer. We found that the model is almost twice as likely to predict a false positive than it is to predict a false negative. This means the model is far more likely to predict a generous tipper and be wrong, than to predict a normal tipper and be wrong. This is less desirable, because it’s better for a driver to be pleasantly surprised by a generous tip when they weren’t expecting one than to be disappointed by a low tip when they were expecting a generous one. However, the overall performance of this model is satisfactory in creating useful predictions.

Lastly, we created a bar graph to plot the most influencial features that affects whether a customer will be a generous tipper or not in a descending order. We found that vendor id(who the driver is), predicted fare(how much the trip itself is expected to cost), average duration (how long the trip last), and average distance plays a heavy impact whether a customer would be a generous tipper or not

Executive Summary

This model performs acceptably. Its F1 score was 0.7235 and it had an overall accuracy of 0.6865. It correctly identified ~78% of the actual responders in the test set, which is 48% better than a random guess.

Unfortunately, random forest is not the most transparent machine learning algorithm. We know that VendorID, predicted fare, mean duration, and mean distance are the most important features, but we don’t know how they influence tipping. This would require further exploration. It is interesting that VendorID is the most predictive feature. This seems to indicate that the drivers themselves are big factors in attracting more generous customers. It may be worth performing statistical tests on the different vendors to examine this further.

It would be very helpful to have past tipping behavior for each customer. It would also be valuable to have accurate tip values for customers who pay with cash, as well as to have a lot more data. With enough data, we could also create a unique feature for each pickup/dropoff combination. This would allow our prediction of generous tippers to be more accurate on location, types of payment, and maybe even general backgrounds on the customers themselves.

With the information and Random Forest Classifier Machine Learning model in hand, however, this could allow TLC to create a better algorithm for their drivers. Letting them understand what creates a good tip, and setting up trips to take advantage of that information gives their drivers an incredible advantage of maximizing their revenue on the road.