Add Anuj Singla Data-OA Submission

.gitignore

@@ -0,0 +1,3 @@
+
+consumer_complaints.csv
+.DS_Store

generate_report.py

@@ -0,0 +1,194 @@
+from fpdf import FPDF
+
+pdf = FPDF()
+WIDTH = 210
+HEIGHT = 297
+X = 0
+Y = 0
+
+def set_letterhead(pdf: FPDF = pdf):
+    pdf.image("./resources/letterhead.png", -1, 0, WIDTH*1.1, HEIGHT*1.18)
+    pass
+
+def set_title_font(pdf: FPDF = pdf):
+    pdf.set_font('Arial', 'B', 22)
+
+def set_heading_1_font(pdf: FPDF = pdf):
+    pdf.set_font('Arial', 'B', 18)
+
+def set_heading_2_font(pdf: FPDF = pdf):
+    pdf.set_font('Arial', 'I', 14)
+
+def set_p_font(pdf: FPDF = pdf):
+    pdf.set_font('Arial', '', 11)
+
+def set_p_i_font(pdf: FPDF = pdf):
+    pdf.set_font('Arial', 'I', 11)
+
+def lb(size: float = 10):
+    global Y
+    Y += size
+    pdf.ln(size)
+
+
+### PAGE 1
+pdf.add_page()
+
+# Letterhead
+set_letterhead()
+
+# Title of report
+set_title_font()
+lb(12)
+pdf.write(5, f"Consumer Financial Complaints Report")
+lb(8)
+
+set_heading_2_font()
+pdf.write(5, "Anuj Singla")
+lb(7)
+
+set_p_font()
+pdf.write(5, "September 28th, 2022")
+lb(15)
+
+# Introduction
+set_heading_1_font()
+pdf.write(10, "Introduction")
+lb()
+
+set_p_font()
+pdf.write(5, ("As a new financial institution entering alpha, our mission is to optimize "
+              "the efficiency of every facet of our operations. Consumer complaints are "
+              "inevitable, but handling disputed resolutions can often become a costly and "
+              "time-consuming process. It is in our best interest to examine our clients' "
+              "backgrounds and prevent them from disputing claims by offering them sufficient "
+              "initial resolutions."))
+lb(15)
+
+# Data
+set_heading_1_font()
+pdf.write(10, "Data")
+lb()
+
+# Data visualization 1
+set_heading_2_font()
+pdf.write(10, "1) How does product affect dispute rate?")
+lb(35)
+
+image_width = WIDTH / 1.1 # Make image fit comfortbly on the page
+image_height = WIDTH / (1.1 * 1132 / 525) # Set height proportional to width and original pixel dimensions
+pdf.image("./resources/visualizations/product-disputes.png", 7, Y, image_width, image_height)
+lb(image_height / 1.4)
+
+set_p_font()
+pdf.write(5,"Mortgages are the most frequent complaint product and have the highest rate of disputes. "
+             "On the other hand, although credit reporting has a fairly large amount of complaints "
+             "(~15,000), it has a lower rate of disputes compared to the rest of the products.")
+lb(6)
+set_p_i_font()
+pdf.write(6, "    -   Why are mortgages disputed so frequently?\n"
+             "    -   Why is credit reporting disputed so rarely?\n"
+             "    -   Why do mortgages have over twice the amount of complaints as any other product?")
+
+
+### PAGE 2
+pdf.add_page()
+Y = 0
+
+# Letterhead
+set_letterhead()
+
+# Data visualization 2
+set_heading_2_font()
+pdf.write(10, "2) How does location affect dispute rate?")
+lb(20)
+
+pdf.image("./resources/visualizations/state-disputes.png", 7, Y, image_width, image_height)
+lb(image_height / 1.3)
+
+set_p_font()
+pdf.write(5, "Consumers in the west coast seem to be more likely to dispute their "
+             "complaint resolutions compared to the rest of the country, followed "
+             "by the east coast. The middle region of the country, highlighted by "
+             "New Mexico and South Dakota, have much lower dispute rates.")
+lb(6)
+set_p_i_font()
+pdf.write(6, "    -   Why does the west coast dispute their complaints so frequently?\n"
+             "    -   Why does the plains region generally have such low dispute rates?")
+lb(15)
+
+# Data visualization 3
+set_heading_2_font()
+pdf.write(10, "3) How does complaint submission method affect dispute rate?")
+lb(35)
+
+pdf.image("./resources/visualizations/submission-disputes.png", 7, Y, image_width, image_height)
+lb(image_height / 1.5)
+
+set_p_font()
+pdf.write(5, "Web submissions, which also happen to be the most common form of complaint "
+             "submission, have the highest dispute rate. On the other hand, postal mail "
+             "submissions, which are one of the least common methods for complaint submissions, "
+             "have the lowest dispute rates. Most notably, instant and non-verbal forms "
+             "of communication (web, fax, email) have higher dispute rates than slower, "
+             "non-anonymous methods of submission (referral, phone, postal mail).")
+lb(6)
+set_p_i_font()
+pdf.write(6, "    -   Do time-consuming submission methods create a barrier for consumers to dispute?")
+
+
+### PAGE 3
+pdf.add_page()
+Y = 0
+
+# Letterhead
+set_letterhead()
+
+# Data visualization 3
+set_heading_2_font()
+pdf.write(10, "4) Sentiment analysis on the most common issues")
+lb(25)
+
+image_width /= 2.15
+image_height /= 2.15
+
+pdf.image("./resources/visualizations/all-complaints-wc.png", 10, Y, image_width, image_height)
+pdf.image("./resources/visualizations/disputed-complaints-wc.png", 20+image_width, Y, image_width, image_height)
+lb(image_height / 1.4)
+
+set_p_i_font()
+pdf.cell(w=image_width, h=6, txt='WordCloud of All Issues', align='C')
+pdf.cell(w=image_width/9, h=6, txt='', align='C')
+pdf.cell(w=image_width, h=6, txt='WordCloud of Disputed Issues Only', align='C')
+lb(10)
+
+set_p_font()
+pdf.write(5, "A form of modification (loan, collection) is the most common issue found "
+"in sentiment analysis. Credit reports seem to be less common among disputes than all "
+"issues. However, servicing payments seems to be an issue that leads to greater disputes.")
+lb(6)
+
+set_p_i_font()
+pdf.write(6, "    -   Why is servicing payments more common among disputed issues?\n"
+             "    -   Why are credit reports less common among disputed issues?\n"
+             "    -   What makes loan and collection modification such common issues?")
+lb(15)
+
+# Introduction
+set_heading_1_font()
+pdf.write(10, "Conclusion")
+lb()
+
+set_p_font()
+pdf.write(5, "Further research and exploration is required to make any concrete conclusions. "
+             "However, these current data insights can orient us in the right direction."
+             "We must gather more data to answer the questions generated from such visualiaitons, "
+             "and in doing so, generate quality financial complaint resolutions for our consumers.")
+lb()
+
+set_p_i_font()
+pdf.write(5, "Note: Future hypotheses are located in the GitHub LaTeX file.")
+
+
+### EXPORT REPORT AS PDF
+pdf.output('report.pdf')

hypotheses.tex

@@ -0,0 +1,43 @@
+\documentclass{report}
+\usepackage[utf8]{inputenc}
+
+\title{Financial Complaints Project Future Hypotheses}
+\author{Anuj Singla}
+\date{September 29th, 2022}
+
+\setcounter{secnumdepth}{3}
+
+\begin{document}
+
+\maketitle
+
+\chapter{Current Project}
+
+\section{Optimizing the ML Model}
+Only one scikit-learn Logistic Regression model was used and was biased toward classifying a complaint in the "non-dispute" category.
+\subsection{If an ensemble method such as XGBoost is optimized for the data set, then the model will make less biased predictions than the Logistic Regression model.}
+\subsection{If the data is balanced to include less non-disputes, then the ML classifier will train with less bias.}
+\subsection{If stronger feature analysis and engineering is performed on the data set, then the ML model will have greater performance with respect to consumer disputes.}
+
+\section{Data Insights}
+The following features were compared to consumer disputes: product, state, submission method, and issue.
+\subsection{If a product is more commonly used within a financial institution, then it will have more complaints.}
+\subsection{If a consumer is located in a state on the western coast of the United States, then they are more likely to dispute their resolution.}
+\subsection{If a consumer uses an online method of submitting their complaint, then they are more likely to dispute their resolution.}
+\subsection{If a consumer submits an issue for servicing payments, then they are more likely to dispute their resolution.}
+
+\section{Additional Dispute Prediction Project Implementations}
+\subsection{If sentiment analysis on issues is used as a feature in the ML Model, then the classifier will have stronger performance.}
+\subsection{If dimensionality reduction with principal component analysis (PCA) is used, then strong, new visualizations and insights can be drawn from the data.}
+
+
+\chapter{Additional Ideas}
+
+\section{Data Exploration}
+\subsection{Interactive map that highlights different zip codes and their attributes.}
+\subsection{Advanced sentiment analysis on product, sub-product, issue, and sub-issue.}
+
+\section{ML Prediction}
+\subsection{Predicting how the company will resolve the complaint}
+
+\end{document}