Goal is to predict Smokers and Drinkers using body signal data.

• Manual Observation in identifying variables and getting familiar with the dataset

Dataset contains medical records of individual between age 20 to 85. The records consist of details of physical examination such as height, weight, sight, hearing, BP and other health signals through blood test of cholesterol, liver function. Kidney etc.

Using health statistic, we are trying to classify individual as smokers (may be they quit smoking regular smokers, chain smokers, etc.) and also check if they are consuming alcohol .

We can use physical examination data to identify outliers values and remove that data.

We can correlate the HDL, LDL, total cholesterol, haemoglobin data values to smokers, and the kidney and liver function tests (SGoT, urine protein etc.) to drinkers.

And based on new test records of an individual we may try to predict what type of smoker he or she would be.

Exploratory Data Analysis (EDA)

EDA purpose is to understand the dataset, cleanse it and analyse the relationship between variable.

Import libraries such as numpy, pandas, matplotlib, seaborn etc. for the analysis.

.shape returns the number of rows by the number of columns for my dataset. My output was (991346, 24), meaning the dataset has 991346rows and 24 columns.

.head() returns the first 5 rows of my dataset. This is useful if you want to see some example values for each variable.

https://github.com/SaarthChahal/ML-DL/blob/main/first%205%20rows%20of%20dataset.png

.columns returns the name of all of your columns in the dataset.

https://github.com/SaarthChahal/ML-DL/blob/main/dataset%20columns.png

After this I worked on getting better understanding of the different values for each variable.

.nunique(axis=0) returns the number of unique values for each variable.

.describe() summarizes the count, mean, standard deviation, min, and max for numeric variables

https://github.com/SaarthChahal/ML-DL/blob/main/describe%20dataset.png

data.describe().apply(lambda s: s.apply(lambda x: format(x, 'f'))) output

Cleaning your DataSet by removing outliers, nulls.

Used .dropna(axis=0) to remove any rows with null values. There was no null value so cleaned data still returned the same (991346, 24) for data_cleaned.shape
Removed outliers by using varaible.between( lower limit, upper limit ) and variable < limit
Data for analysis does not use string datatype as an argument , I could have encoded the gender variable labelled as ‘sex’ to number for male or female or I could have dropped column itself. I choose to drop the column by using .drop(‘variable’)
Also I encoded /converted DRK_YN variable string values ( Y on N) to number 1 or 0

https://github.com/SaarthChahal/ML-DL/blob/main/cleaned%20dataset.png

From shapes output: (985543, 23) i.e I was able to reduce 5803 records and 1 column.

Data Plotting exercise to analyze relation ship between variables. Calculate the correlation matrix.
There are too many variables to produce more readable correlation matrix and heatmap. Created 2 smaller array for matrix and heatmap for smoke and drink correlation

Heatmap for smokers
https://github.com/SaarthChahal/ML-DL/blob/main/heatmap%20for%20smokers.png

Heatmap for drinkers
https://github.com/SaarthChahal/ML-DL/blob/main/heatmap%20for%20drinkers.png

Scatterplot for total cholestrol of smokers
https://github.com/SaarthChahal/ML-DL/blob/main/total%20cholestrol%20scatterplot%20for%20smokers.png

using sns.pairplot() created scatterplots between some of key variables
https://github.com/SaarthChahal/ML-DL/blob/main/scatterplots%20for%20pairs%20of%20variables.png

Model training module

LINEAR REGRESSION MODEL

Train linear regression model.

We will need to first split up our data into an X1 array(cholesterol) that contains the features to train on,
And a y1 array(SMK_stat_type_cd) with the target variable,
split up our data into an X2 array(Kidney function) that contains the features to train on,
And a y2 array(DRK_YN)

Train test split. test split is 40 % train set is 60 %

Loading the linear regression Model
prediction on Training data

Model evlauation.
Let's evaluate the model by checking out it's coefficients and how we can interpret them.

Learning model intercept 1 (smokers): -1.2726375396245835

Learning model intercept 2 (drinkers) : 0.4141733052412185
Coefficients for smokers: https://github.com/SaarthChahal/ML-DL/blob/main/coefficient.png

Coefficient for drinkers: https://github.com/SaarthChahal/ML-DL/blob/main/coefficient2.png

Interpreting the coefficient.

For every one unit change in smoke status there is negative impact on Cholestrol ( refelcted as negative)

and increase in triglyceride and hemoglobin which negatively affect the health indicator.

Prediction from Model
Prediction scatterplot for smokers: https://github.com/SaarthChahal/ML-DL/blob/main/prediction%20scatterplot%20for%20smokers.png

Displot prediction for smokers: https://github.com/SaarthChahal/ML-DL/blob/main/displot%20method%20for%20smokers.png

Prediction scatterplot for drinkers: https://github.com/SaarthChahal/ML-DL/blob/main/scatterplot%20prediction%20for%20drinkers.png

Displot scatterplot for drinkers: https://github.com/SaarthChahal/ML-DL/blob/main/displot%20method%20for%20smokers.png

Regression Evaluation Metrics
Here are three common evaluation metrics for regression problems:

Mean Absolute Error** (MAE) is the mean of the absolute value of the errors: is the easiest to understand, because it's the average error.

Mean Squared Error** (MSE) is the mean of the squared errors: is more popular than MAE, because MSE "punishes" larger errors, which tends to be useful in the real world.

Root Mean Squared Error** (RMSE) is the square root of the mean of the squared errors: is even more popular than MSE, because RMSE is interpretable in the "y" units.

Regression Evaluation Metrics for smokers:

MAE:1 1.1094765424193946

MSE:1 1.8638480376557032

RMSE:1 1.3652281998463491

Regression Evaluation Metrics for drinkers:

MAE:2 0.4814403810202582

MSE:2 0.23847155682695156

RMSE:2 0.4883354961775271

List of variables used: https://github.com/SaarthChahal/ML-DL/blob/main/variables.png

DECISION TREE AND RANDOM FOREST MODEL

Supervised, regression machine learning problem. It’s supervised because we have both the features (data on health parameters) and the targets (Smokers and Drinkers) that we want to predict

The reported averages include macro average (averaging the unweighted mean per label), weighted average (averaging the support-weighted mean per label), and sample average (only for multilabel classification). Micro average (averaging the total true positives, false negatives and false positives) is only shown for multi-label or multi-class with a subset of classes, because it corresponds to accuracy otherwise and would be the same for all metrics

Classfication metrics.

The precision is the ratio tp / (tp + fp) where tp is the number of true positives and fp the number of false positives. The precision is intuitively the ability of the classifier not to label a negative sample as positive.

The recall is the ratio tp / (tp + fn) where tp is the number of true positives and fn the number of false negatives. The recall is intuitively the ability of the classifier to find all the positive samples.

Fscore. The F-beta score can be interpreted as a weighted harmonic mean of the precision and recall, where an F-beta score reaches its best value at 1 and worst score at 0.
The F-beta score weights recall more than precision by a factor of beta. beta == 1.0 means recall and precision are equally important.

Smokers metrics for decision tree:

          precision    recall   f1-score  support

       1       0.77      0.75      0.76    179903
       2       0.31      0.32      0.32     52132
       3       0.40      0.41      0.40     63628

accuracy                           0.60    295663
macro avg      0.49      0.49      0.49    295663
weighted avg   0.61      0.60      0.61    295663

Smokers metrics for random forest model:

           precision   recall   f1-score  support

       1       0.80      0.85      0.82    179903
       2       0.43      0.32      0.37     52132
       3       0.52      0.54      0.53     63628

accuracy                           0.69    295663
macro avg       0.58      0.57     0.57    295663
weighted avg    0.67      0.69     0.68    295663

Drinkers metrics for decision tree:

             precision    recall  f1-score   support

       0       0.63      0.63      0.63    147705
       1       0.63      0.63      0.63    147958

accuracy                           0.63    295663
macro avg       0.63      0.63     0.63    295663
weighted avg    0.63      0.63     0.63    295663

Drinkers metrics for random forest model:

             precision    recall  f1-score   support

       0       0.72       0.72      0.72    147705
       1       0.72       0.72      0.72    147958

accuracy                            0.63    295663
macro avg       0.72      0.72      0.63    295663
weighted avg    0.72      0.72      0.63    295663

Confusion Matrix.

Confusion matrix to evaluate the accuracy of a classification.

Confusion matrix usage to evaluate the quality of the output of a classifier on the data.

The diagonal elements represent the number of points for which the predicted label is equal to the true label, while off-diagonal elements are those that are mislabeled by the classifier.

The higher the diagonal values of the confusion matrix the better, indicating many correct predictions.

Decision Tree Confusion Matrix for smokers:

[[135735  20884  23284]

[ 19324  16859  15949]

[ 21749  15999  25880]]

Random Forest Confusion Matrix for smokers:

[[152378  11722  15803]

[ 19136  16504  16492]

[ 19491   9826  34311]]

Decision Tree Confusion Matrix for drinkers:

[[93021 54684]
[54020 93938]]

Random Forest Confusion Matrix for drinkers:

[[106993  40712]
[ 41075 106883]]

Conclusion

To compare linear regression, decision trees, and random forests for supervised learning, let's assess them based on performance, accuracy, computational speed, and memory usage. Keep in mind that the specific results may vary depending on your dataset and implementation, but I can provide a general comparison:

Linear Regression:

• Performance:
  Linear regression performs well when the relationship between features and the target variable is roughly linear.
• Accuracy:
  Accuracy is moderate when dealing with linear relationships but may suffer when data has complex, non-linear patterns.
• Computational Speed:
  Linear regression is very computationally efficient and quick to train.
• Memory Usage:
  Linear regression models have a small memory footprint as they store coefficients for each feature.

Decision Tree:

• Performance:
  Decision trees are versatile and can capture both linear and non-linear relationships in data.
• Accuracy:
  Decision trees can provide high accuracy but may overfit when not properly pruned.
• Computational Speed:
  Building decision trees is usually fast, especially for small to moderately sized datasets.
• Memory Usage:
  Decision trees can consume moderate memory, especially if they are deep.

Random Forest:

• Performance:
  Random forests are robust and handle non-linearity well, making them suitable for a wide range of problems.
• Accuracy:
  Random forests often provide high accuracy and reduce overfitting through ensemble techniques.
• Computational Speed:
  Training a random forest involves building multiple decision trees, which can be slower than linear regression but faster than training a deep single decision tree for large datasets.
• Memory Usage:
  Random forests consume more memory than linear regression but are typically more memory-efficient than deep decision trees.

In summary:

Linear Regression is computationally efficient and interpretable but may not perform well with complex, non-linear data.

Decision Trees are versatile and can capture non-linear relationships but may overfit and require careful pruning.

Random Forests are robust, accurate, and handle non-linearity well, making them suitable for a wide range of tasks. They offer a balance between accuracy and computational efficiency.

Recommendations

I would recommend using the Random Forest model as it provides the highest accuracy, the computational speed is slower in comparison but speed is not crucial for analyzing data on smokers and drinkers.