Health Costs Predictor Model

python
Published

January 1, 2024

import matplotlib.pyplot as plt
import pandas as pd

# 2.x
from tensorflow import keras
from tensorflow.keras import layers
import requests

url = "https://cdn.freecodecamp.org/project-data/health-costs/insurance.csv"
query_parameters = {"downloadformat": "csv"}

response = requests.get(url, params=query_parameters)

with open("insurance.csv", mode="wb") as file:
    file.write(response.content)
dataset = pd.read_csv("insurance.csv")
dataset.tail()
age sex bmi children smoker region expenses
1333 50 male 31.0 3 no northwest 10600.55
1334 18 female 31.9 0 no northeast 2205.98
1335 18 female 36.9 0 no southeast 1629.83
1336 21 female 25.8 0 no southwest 2007.95
1337 61 female 29.1 0 yes northwest 29141.36 
dataset.isna().sum()

# region_list = dataset["region"].unique()
# region_map = {i: e for e, i in enumerate(region_list)}
# region_map
age         0
sex         0
bmi         0
children    0
smoker      0
region      0
expenses    0
dtype: int64
# dataset["sex"] = dataset["sex"].map({"male": 1, "female": 2})
# dataset["smoker"] = dataset["smoker"].map({"yes": 1, "no": 0})
# dataset["region"] = dataset["sex"].map(region_map)

dataset = pd.get_dummies(dataset, columns=["sex"], prefix="sex", prefix_sep="_")
dataset = pd.get_dummies(dataset, columns=["smoker"], prefix="smoker", prefix_sep="_")
dataset = pd.get_dummies(dataset, columns=["region"], prefix="region", prefix_sep="_")

dataset.tail()
age bmi children expenses sex_female sex_male smoker_no smoker_yes region_northeast region_northwest region_southeast region_southwest
1333 50 31.0 3 10600.55 False True True False False True False False
1334 18 31.9 0 2205.98 True False True False True False False False
1335 18 36.9 0 1629.83 True False True False False False True False
1336 21 25.8 0 2007.95 True False True False False False False True
1337 61 29.1 0 29141.36 True False False True False True False False 
from sklearn.model_selection import train_test_split

train_dataset, test_dataset = train_test_split(dataset, test_size=0.2, random_state=0)
train_features = train_dataset.copy()
test_features = test_dataset.copy()

train_labels = train_features.pop("expenses")
test_labels = test_features.pop("expenses")
model = keras.Sequential(
    [
        layers.Dense(units=16, activation="relu"),
        layers.Dense(units=8, activation="relu"),
        layers.Dense(units=1, activation="linear"),
    ]
)

model.summary()
Model: "sequential"

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ dense (Dense)                   │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_1 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_2 (Dense)                 │ ?                      │   0 (unbuilt) │
└─────────────────────────────────┴────────────────────────┴───────────────┘

 Total params: 0 (0.00 B)

 Trainable params: 0 (0.00 B)

 Non-trainable params: 0 (0.00 B)

model.compile(
    optimizer=keras.optimizers.Adam(learning_rate=0.1),
    loss="mean_absolute_error",
    metrics=["mae"],
)
history = model.fit(
    train_features,
    train_labels,
    epochs=50,
    verbose=1,
    validation_split=0.2,
)
Epoch 1/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 2s 12ms/step - loss: 12400.3613 - mae: 12400.3613 - val_loss: 8794.2754 - val_mae: 8794.2754
Epoch 2/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 7730.8154 - mae: 7730.8154 - val_loss: 8197.8213 - val_mae: 8197.8213
Epoch 3/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 6926.5454 - mae: 6926.5454 - val_loss: 7878.8511 - val_mae: 7878.8511
Epoch 4/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 6531.6499 - mae: 6531.6499 - val_loss: 7495.9204 - val_mae: 7495.9204
Epoch 5/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 5858.8516 - mae: 5858.8516 - val_loss: 7229.2598 - val_mae: 7229.2598
Epoch 6/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 6184.5981 - mae: 6184.5981 - val_loss: 7193.7910 - val_mae: 7193.7910
Epoch 7/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 6406.2319 - mae: 6406.2319 - val_loss: 6683.9468 - val_mae: 6683.9468
Epoch 8/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 4873.6665 - mae: 4873.6665 - val_loss: 6323.3506 - val_mae: 6323.3506
Epoch 9/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 5404.1587 - mae: 5404.1587 - val_loss: 5840.2300 - val_mae: 5840.2300
Epoch 10/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 4306.5776 - mae: 4306.5776 - val_loss: 4339.7681 - val_mae: 4339.7681
Epoch 11/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step - loss: 3330.8904 - mae: 3330.8904 - val_loss: 4155.7017 - val_mae: 4155.7017
Epoch 12/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 3358.3098 - mae: 3358.3098 - val_loss: 4099.0825 - val_mae: 4099.0825
Epoch 13/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step - loss: 3719.3210 - mae: 3719.3210 - val_loss: 4288.0825 - val_mae: 4288.0825
Epoch 14/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 3741.3633 - mae: 3741.3633 - val_loss: 4175.4727 - val_mae: 4175.4727
Epoch 15/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 3244.9753 - mae: 3244.9753 - val_loss: 3973.3491 - val_mae: 3973.3491
Epoch 16/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step - loss: 2959.1372 - mae: 2959.1372 - val_loss: 3972.8040 - val_mae: 3972.8040
Epoch 17/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 3164.6609 - mae: 3164.6609 - val_loss: 3850.2273 - val_mae: 3850.2273
Epoch 18/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2713.0781 - mae: 2713.0781 - val_loss: 3795.8066 - val_mae: 3795.8066
Epoch 19/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2997.2271 - mae: 2997.2271 - val_loss: 3859.2603 - val_mae: 3859.2603
Epoch 20/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 3085.3821 - mae: 3085.3821 - val_loss: 3639.3989 - val_mae: 3639.3989
Epoch 21/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2957.4858 - mae: 2957.4858 - val_loss: 3621.8835 - val_mae: 3621.8835
Epoch 22/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step - loss: 2837.3000 - mae: 2837.3000 - val_loss: 3489.7847 - val_mae: 3489.7847
Epoch 23/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2905.4102 - mae: 2905.4102 - val_loss: 3787.1067 - val_mae: 3787.1067
Epoch 24/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step - loss: 3141.4758 - mae: 3141.4758 - val_loss: 3531.4854 - val_mae: 3531.4854
Epoch 25/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2478.5315 - mae: 2478.5315 - val_loss: 3507.5911 - val_mae: 3507.5911
Epoch 26/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2816.0601 - mae: 2816.0601 - val_loss: 3532.2092 - val_mae: 3532.2092
Epoch 27/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2678.4246 - mae: 2678.4246 - val_loss: 3658.4497 - val_mae: 3658.4497
Epoch 28/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 2833.5803 - mae: 2833.5803 - val_loss: 3488.4082 - val_mae: 3488.4082
Epoch 29/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2742.7410 - mae: 2742.7410 - val_loss: 3464.1750 - val_mae: 3464.1750
Epoch 30/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2677.0469 - mae: 2677.0469 - val_loss: 3482.8792 - val_mae: 3482.8792
Epoch 31/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 2ms/step - loss: 2569.9336 - mae: 2569.9336 - val_loss: 3353.8123 - val_mae: 3353.8123
Epoch 32/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2737.5840 - mae: 2737.5840 - val_loss: 3485.8455 - val_mae: 3485.8455
Epoch 33/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2512.9038 - mae: 2512.9038 - val_loss: 3309.5256 - val_mae: 3309.5256
Epoch 34/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2351.0886 - mae: 2351.0886 - val_loss: 3229.2351 - val_mae: 3229.2351
Epoch 35/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2360.1023 - mae: 2360.1023 - val_loss: 3396.7778 - val_mae: 3396.7778
Epoch 36/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 2430.5044 - mae: 2430.5044 - val_loss: 3259.0562 - val_mae: 3259.0562
Epoch 37/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2512.6514 - mae: 2512.6514 - val_loss: 3160.6106 - val_mae: 3160.6106
Epoch 38/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 2388.4092 - mae: 2388.4092 - val_loss: 3180.7351 - val_mae: 3180.7351
Epoch 39/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2303.3711 - mae: 2303.3711 - val_loss: 3262.7922 - val_mae: 3262.7922
Epoch 40/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2160.1658 - mae: 2160.1658 - val_loss: 3144.8740 - val_mae: 3144.8740
Epoch 41/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2304.2559 - mae: 2304.2559 - val_loss: 3094.5076 - val_mae: 3094.5076
Epoch 42/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2093.5830 - mae: 2093.5830 - val_loss: 3093.9961 - val_mae: 3093.9961
Epoch 43/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2485.9470 - mae: 2485.9470 - val_loss: 3144.2881 - val_mae: 3144.2881
Epoch 44/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2476.5474 - mae: 2476.5474 - val_loss: 3024.2004 - val_mae: 3024.2004
Epoch 45/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2113.4941 - mae: 2113.4941 - val_loss: 3064.6553 - val_mae: 3064.6553
Epoch 46/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2147.3889 - mae: 2147.3889 - val_loss: 3155.6006 - val_mae: 3155.6006
Epoch 47/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2598.4773 - mae: 2598.4773 - val_loss: 3005.9568 - val_mae: 3005.9568
Epoch 48/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2133.7805 - mae: 2133.7805 - val_loss: 2994.0830 - val_mae: 2994.0830
Epoch 49/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 4ms/step - loss: 2203.0266 - mae: 2203.0266 - val_loss: 2930.2429 - val_mae: 2930.2429
Epoch 50/50
27/27 ━━━━━━━━━━━━━━━━━━━━ 0s 3ms/step - loss: 2550.6213 - mae: 2550.6213 - val_loss: 3008.1443 - val_mae: 3008.1443
def plot_loss(history):
    plt.plot(history.history["loss"], label="loss")
    plt.plot(history.history["val_loss"], label="val_loss")
    plt.ylim([0, 15_000])
    plt.xlabel("Epoch")
    plt.ylabel("Error")
    plt.legend()
    plt.grid(True)


plot_loss(history)

# Renaming for the test
test_dataset = test_features

# RUN THIS CELL TO TEST YOUR MODEL. DO NOT MODIFY CONTENTS.
# Test model by checking how well the model generalizes using the test set.
# loss, mae, mse = model.evaluate(test_dataset, test_labels, verbose=2)
loss, mae = model.evaluate(test_dataset, test_labels, verbose=2)
# mae = model.evaluate(test_dataset, test_labels, verbose=2)

print("Testing set Mean Abs Error: {:5.2f} expenses".format(mae))

if mae < 3500:
    print("You passed the challenge. Great job!")
else:
    print("The Mean Abs Error must be less than 3500. Keep trying.")

# Plot predictions.
test_predictions = model.predict(test_dataset).flatten()

a = plt.axes(aspect="equal")
plt.scatter(test_labels, test_predictions)
plt.xlabel("True values (expenses)")
plt.ylabel("Predictions (expenses)")
lims = [0, 50000]
plt.xlim(lims)
plt.ylim(lims)
_ = plt.plot(lims, lims)
9/9 - 0s - 6ms/step - loss: 2247.7083 - mae: 2247.7083
Testing set Mean Abs Error: 2247.71 expenses
You passed the challenge. Great job!
9/9 ━━━━━━━━━━━━━━━━━━━━ 0s 6ms/step

Created in deepnote.com Created in Deepnote