Prabuddha21/encrypted_credit_scoring
0
1"""A gradio app for credit card approval prediction using FHE."""2 3import subprocess4import time5import gradio as gr6 7from settings import (8 REPO_DIR,9 ACCOUNT_MIN_MAX,10 CHILDREN_MIN_MAX,11 INCOME_MIN_MAX,12 AGE_MIN_MAX,13 FAMILY_MIN_MAX,14 INCOME_TYPES,15 OCCUPATION_TYPES,16 HOUSING_TYPES,17 EDUCATION_TYPES,18 FAMILY_STATUS,19 YEARS_EMPLOYED_BINS,20 INCOME_VALUE,21 AGE_VALUE,22)23from backend import (24 keygen_send,25 pre_process_encrypt_send_applicant,26 pre_process_encrypt_send_bank,27 pre_process_encrypt_send_credit_bureau,28 run_fhe,29 get_output_and_decrypt,30 explain_encrypt_run_decrypt,31)32 33 34subprocess.Popen(["uvicorn", "server:app"], cwd=REPO_DIR)35time.sleep(3)36 37 38demo = gr.Blocks()39 40 41print("Starting the demo...")42with demo:43 44 # gr.Markdown(45 # """46 # <p align="center">47 # <img width=200 src="https://user-images.githubusercontent.com/5758427/197816413-d9cddad3-ba38-4793-847d-120975e1da11.png">48 # </p>49 # """50 # )51 # gr.Markdown(52 # """53 # <h1 align="center">Encrypted Credit Card Approval Prediction Using Fully Homomorphic Encryption</h1>54 # <p align="center">55 # <a href="https://github.com/zama-ai/concrete-ml"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="file/images/logos/github.png">Concrete-ML</a>56 # —57 # <a href="https://docs.zama.ai/concrete-ml"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="file/images/logos/documentation.png">Documentation</a>58 # —59 # <a href="https://zama.ai/community"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="file/images/logos/community.png">Community</a>60 # —61 # <a href="https://twitter.com/zama_fhe"> <img style="vertical-align: middle; display:inline-block; margin-right: 3px;" width=15 src="file/images/logos/x.png">@zama_fhe</a>62 # </p>63 # """64 # )65 66 with gr.Accordion("What is credit scoring for card approval?", open=False):67 gr.Markdown(68 """69 It is a complex process that involves several entities: the applicant, the bank, the 70 credit bureau, and the credit scoring agency. When you apply for a credit card, you71 provide personal and financial information to the bank. This might include your income,72 employment status, and existing debts. The bank uses this information to assess your 73 creditworthiness. To do this, they often turn to credit bureaus and credit scoring 74 agencies. 75 - Credit bureaus collect and maintain data on consumers' credit and payment 76 histories. This data includes your past and current debts, payment history, and the 77 length of your credit history. 78 - Credit scoring agencies use algorithms to analyze 79 the data from credit bureaus and generate a credit score. This score is a numerical 80 representation of your creditworthiness. 81 - The bank uses your credit score, along with 82 the information you provided, to make a decision on your credit card application. A 83 higher credit score generally increases your chances of being approved and may result 84 in better terms (like a lower interest rate). 85 """86 )87 88 with gr.Accordion("Why is it critical to add a new privacy layer to this process?", open=False):89 gr.Markdown(90 """91 The data involved is highly sensitive. It includes personal details like your Social 92 Security number, income, and credit history. There's significant sharing of data 93 between different entities. Your information is not just with the bank, but also with 94 credit bureaus and scoring agencies. The more entities that have access to your data, 95 the greater the risk of a data breach. This can lead to identity theft and financial 96 fraud. There's also the issue of data accuracy. Mistakes in credit reports can lead to 97 unjustly low credit scores, affecting your ability to get credit. 98 """99 )100 101 with gr.Accordion(102 "Why is Fully Homomorphic Encryption (FHE) a solution for better credit scoring?", 103 open=False,104 ):105 gr.Markdown(106 """107 Fully Homomorphic Encryption (FHE) is seen as an ideal solution for enhancing privacy 108 and accuracy in credit scoring processes involving multiple parties like applicants, 109 banks, credit bureaus, and credit scoring agencies. It allows data to be encrypted and 110 processed without ever needing to decrypt it. This means that sensitive data can be 111 shared and analyzed without exposing the actual information to any of the parties or 112 the server processing it. In the context of credit scoring, this would enable a more 113 thorough and accurate assessment of a person's creditworthiness. Data from various 114 sources can be combined and analyzed to make a more informed decision, yet each party's 115 data remains confidential. As a result, the risk of data leaks or breaches is 116 significantly minimized, addressing major privacy concerns. 117 118 To summarize, FHE provides a means to make more accurate credit eligibility decisions 119 while maintaining strict data privacy, offering a sophisticated solution to the delicate 120 balance between data utility and confidentiality.121 """122 )123 124 gr.Markdown(125 """126 <p align="center">127 <img src="https://raw.githubusercontent.com/kcelia/Img/main/credit_scoring_banner.png"128 </p>129 """130 )131 132 gr.Markdown("## Step 1: Generate the keys.")133 gr.Markdown("<hr />")134 gr.Markdown("<span style='color:grey'>Applicant, Bank and Credit bureau setup</span>")135 gr.Markdown(136 """137 - The private key is generated jointly by the entities that collaborate to compute the 138 credit score. It is used to encrypt and decrypt the data and shall never be shared with 139 any other party.140 - The evaluation key is a public key that the server needs to process encrypted data. It is141 therefore transmitted to the server for further processing as well.142 """143 )144 keygen_button = gr.Button("Generate the keys and send evaluation key to the server.")145 evaluation_key = gr.Textbox(146 label="Evaluation key representation:", max_lines=2, interactive=False147 )148 client_id = gr.Textbox(label="", max_lines=2, interactive=False, visible=False)149 150 # Button generate the keys151 keygen_button.click(152 keygen_send,153 outputs=[client_id, evaluation_key, keygen_button],154 )155 156 gr.Markdown("## Step 2: Fill in some information.")157 gr.Markdown("<hr />")158 gr.Markdown("<span style='color:grey'>Applicant, Bank and Credit bureau setup</span>")159 gr.Markdown(160 """161 Select the information that corresponds to the profile you want to evaluate. Three sources 162 of information are represented in this model:163 - the applicant's personal information in order to evaluate his/her credit card eligibility;164 - the applicant bank account history, which provides any type of information on the 165 applicant's banking information relevant to the decision (here, we consider duration of 166 account);167 - and credit bureau information, which represents any other information (here, 168 employment history) that could provide additional insight relevant to the decision.169 170 Please always encrypt and send the values (through the buttons on the right) once updated171 before running the FHE inference.172 """173 )174 175 with gr.Row():176 with gr.Column():177 gr.Markdown("### Step 2.1 - Applicant information 🧑💻")178 bool_inputs = gr.CheckboxGroup(179 ["Car", "Property", "Mobile phone"], 180 label="Which of the following do you actively hold or own?"181 )182 num_children = gr.Slider(183 **CHILDREN_MIN_MAX, 184 step=1, 185 label="Number of children", 186 info="How many children do you have ?"187 )188 household_size = gr.Slider(189 **FAMILY_MIN_MAX, 190 step=1, 191 label="Household size", 192 info="How many members does your household have ?"193 )194 total_income = gr.Slider(195 **INCOME_MIN_MAX,196 value=INCOME_VALUE,197 label="Income", 198 info="What's you total yearly income (in euros) ?"199 )200 age = gr.Slider(201 **AGE_MIN_MAX,202 value=AGE_VALUE, 203 step=1, 204 label="Age", 205 info="How old are you ?"206 )207 208 with gr.Column():209 income_type = gr.Dropdown(210 choices=INCOME_TYPES, 211 value=INCOME_TYPES[0], 212 label="Income type", 213 info="What is your main type of income ?"214 )215 education_type = gr.Dropdown(216 choices=EDUCATION_TYPES, 217 value=EDUCATION_TYPES[0], 218 label="Education", 219 info="What is your education background ?"220 )221 family_status = gr.Dropdown(222 choices=FAMILY_STATUS, 223 value=FAMILY_STATUS[0], 224 label="Family", 225 info="What is your family status ?"226 )227 occupation_type = gr.Dropdown(228 choices=OCCUPATION_TYPES, 229 value=OCCUPATION_TYPES[0], 230 label="Occupation", 231 info="What is your main occupation ?"232 )233 housing_type = gr.Dropdown(234 choices=HOUSING_TYPES, 235 value=HOUSING_TYPES[0], 236 label="Housing", 237 info="In what type of housing do you live ?"238 )239 240 with gr.Row():241 with gr.Column(scale=2):242 encrypt_button_applicant = gr.Button("Encrypt the inputs and send to server.")243 244 encrypted_input_applicant = gr.Textbox(245 label="Encrypted input representation:", max_lines=2, interactive=False246 )247 248 gr.Markdown("<hr />")249 with gr.Column():250 gr.Markdown("### Step 2.2 - Bank information 🏦")251 account_age = gr.Slider(252 **ACCOUNT_MIN_MAX, 253 step=1, 254 label="Account age (months)", 255 info="How long have this person had this bank account (in months) ?"256 )257 258 with gr.Row():259 with gr.Column(scale=2):260 encrypt_button_bank = gr.Button("Encrypt the inputs and send to server.")261 262 encrypted_input_bank = gr.Textbox(263 label="Encrypted input representation:", max_lines=2, interactive=False264 )265 266 gr.Markdown("<hr />")267 with gr.Column():268 gr.Markdown("### Step 2.3 - Credit bureau information 🏢")269 employed = gr.Radio(["Yes", "No"], label="Is the person employed ?", value="Yes")270 years_employed = gr.Dropdown(271 choices=YEARS_EMPLOYED_BINS, 272 value=YEARS_EMPLOYED_BINS[0], 273 label="Years of employment", 274 info="How long have this person been employed (in years) ?"275 )276 277 with gr.Row():278 with gr.Column(scale=2):279 encrypt_button_credit_bureau = gr.Button("Encrypt the inputs and send to server.")280 281 encrypted_input_credit_bureau = gr.Textbox(282 label="Encrypted input representation:", max_lines=2, interactive=False283 )284 285 # Button to pre-process, generate the key, encrypt and send the applicant inputs from the client 286 # side to the server287 encrypt_button_applicant.click(288 pre_process_encrypt_send_applicant,289 inputs=[client_id, bool_inputs, num_children, household_size, total_income, age, \290 income_type, education_type, family_status, occupation_type, housing_type],291 outputs=[encrypted_input_applicant, encrypt_button_applicant],292 )293 294 # Button to pre-process, generate the key, encrypt and send the bank inputs from the client 295 # side to the server296 encrypt_button_bank.click(297 pre_process_encrypt_send_bank,298 inputs=[client_id, account_age],299 outputs=[encrypted_input_bank, encrypt_button_bank],300 )301 302 # Button to pre-process, generate the key, encrypt and send the credit bureau inputs from the 303 # client side to the server 304 encrypt_button_credit_bureau.click(305 pre_process_encrypt_send_credit_bureau,306 inputs=[client_id, years_employed, employed],307 outputs=[encrypted_input_credit_bureau, encrypt_button_credit_bureau],308 )309 310 gr.Markdown("## Step 3: Run the FHE evaluation.")311 gr.Markdown("<hr />")312 gr.Markdown("<span style='color:grey'>Server Side</span>")313 gr.Markdown(314 """315 Once the server receives the encrypted inputs, it can compute the prediction without ever 316 needing to decrypt any value.317 318 This server employs a [Decision Tree](https://scikit-learn.org/stable/modules/generated/sklearn.tree.DecisionTreeClassifier.html)319 classifier model that has been trained on a synthetic data-set.320 """321 )322 323 execute_fhe_button = gr.Button("Run the FHE evaluation.")324 fhe_execution_time = gr.Textbox(325 label="Total FHE execution time (in seconds):", max_lines=1, interactive=False326 )327 328 # Button to send the encodings to the server using post method329 execute_fhe_button.click(run_fhe, inputs=[client_id], outputs=[fhe_execution_time, execute_fhe_button])330 331 gr.Markdown("## Step 4: Receive the encrypted output from the server and decrypt.")332 gr.Markdown("<hr />")333 gr.Markdown("<span style='color:grey'>Applicant, Bank and Credit bureau decryption</span>")334 gr.Markdown(335 """336 Once the server completed the inference, the encrypted output is returned to the applicant.337 338 The three entities that provide the information to compute the credit score are the only 339 ones that can decrypt the result. They take part in a decryption protocol that allows to 340 only decrypt the full result when all three parties decrypt their share of the result.341 """342 )343 gr.Markdown(344 """345 The first value displayed below is a shortened byte representation of the actual encrypted 346 output.347 The applicant is then able to decrypt the value using its private key.348 """349 )350 351 get_output_button = gr.Button("Receive the encrypted output from the server.")352 encrypted_output_representation = gr.Textbox(353 label="Encrypted output representation: ", max_lines=2, interactive=False354 )355 prediction_output = gr.Textbox(356 label="Prediction", max_lines=1, interactive=False357 )358 359 # Button to send the encodings to the server using post method360 get_output_button.click(361 get_output_and_decrypt, 362 inputs=[client_id], 363 outputs=[prediction_output, encrypted_output_representation, get_output_button],364 )365 366 gr.Markdown("## Step 5: Explain the prediction (only if your credit card is likely to be denied).")367 gr.Markdown("<hr />")368 gr.Markdown(369 """370 In case the credit card is likely to be denied, the applicant can ask for how many years of 371 employment would most likely be required in order to increase the chance of getting a 372 credit card approval.373 374 All of the above steps are combined into a single button for simplicity. The following 375 button therefore encrypts the same inputs (except the years of employment, which varies) 376 from all three parties, runs the new prediction in FHE and decrypts the output. 377 378 In case the following states to try a new "Years of employment" input, one can simply 379 update the value in Step 2 and directly run Step 6 once more. 380 """381 )382 explain_button = gr.Button(383 "Encrypt the inputs, compute in FHE and decrypt the output."384 )385 explain_prediction = gr.Textbox(386 label="Additional years of employed required.", interactive=False387 )388 389 # Button to explain the prediction390 explain_button.click(391 explain_encrypt_run_decrypt,392 inputs=[client_id, prediction_output, years_employed, employed],393 outputs=[explain_prediction, explain_button],394 )395 396 gr.Markdown(397 "The app was built with [Concrete-ML](https://github.com/zama-ai/concrete-ml), a "398 "Privacy-Preserving Machine Learning (PPML) open-source set of tools by [Zama](https://zama.ai/). "399 "Try it yourself and don't forget to star on Github ⭐."400 )401 402demo.launch(share=False)403