The Support Vector Machines model was built using all of the variables in the prepped dataset.

The following variables were used:

• total_line
• temp
• wind
• total_qb_elo
• team_elo_diff
• qb_elo_diff
• avg_home_total_yards
• avg_away_total_yards
• avg_home_total_yards_against
• avg_away_total_yards_against
• home_away_yards_diff
• away_home_yards_diff
• total_result

C is the regularization parameter for Support Vector Machines. The default for the regularization parameter in sklearn is 1.0. A regularization parameter of 1.0, 1.5 ,and 2.0 will be used for each kernel to determine if changing the parameter will lead to an improved accuracy.

Linear Kernel (C=1.0)

Evaluation Metrics:

• Accuracy: 0.511

The linear kernel with a regularization parameter of 1.0 performed poorly with an accuracy of 51.1%. This is above below the necessary 52.4% to be profitable.

Linear Kernel (C=1.5)

Evaluation Metrics:

• Accuracy: 0.506

The linear kernel with a regularization parameter of 1.5 performed poorly with an accuracy of 50.6%. This is above below the necessary 52.4% to be profitable.

Linear Kernel (C=2.0)

Evaluation Metrics:

• Accuracy: 0.505

The linear kernel with a regularization parameter of 2.0 performed poorly with an accuracy of 50.5%. This is above below the necessary 52.4% to be profitable. Further, it did not predict any totals over the projected total correctly, which is not ideal.

Sigmoid Kernel (C=1.0)

Evaluation Metrics:

• Accuracy: 0.506

The sigmoid kernel with a regularization parameter of 1.0, performed poorly with an accuracy of 50.6%. This is below the necessary 52.4% to be profitable.

Sigmoid Kernel (C=1.5)

Evaluation Metrics:

• Accuracy: 0.487

The sigmoid kernel with a regularization parameter of 1.5, performed poorly with an accuracy of 48.7%. This is below the necessary 52.4% to be profitable.

Sigmoid Kernel (C=2.0)

Evaluation Metrics:

• Accuracy: 0.486

The sigmoid kernel with a regularization parameter of 2.0, performed poorly with an accuracy of 48.6%. This is below the necessary 52.4% to be profitable.

RBF (Gaussian) Kernel (C=1.0)

Evaluation Metrics:

• Accuracy: 0.523

The rbf kernel with a regularization parameter of 1.0, performed poorly with an accuracy of 52.3%. This is below the necessary 52.4% to be profitable.

RBF (Gaussian) Kernel (C=1.5)

Evaluation Metrics:

• Accuracy: 0.522

The rbf kernel with a regularization parameter of 1.5, performed poorly with an accuracy of 52.2%. This is below the necessary 52.4% to be profitable.

RBF (Gaussian) Kernel (C=2.0)

Evaluation Metrics:

• Accuracy: 0.490

The rbf kernel with a regularization parameter of 2.0, performed poorly with an accuracy of 49.0%. This is below the necessary 52.4% to be profitable.

Polynomial Kernel (degree = 2)

Evaluation Metrics:

Accuracy: 0.521

The polynomial kernel with a degree of two, almost performed well enough with an accuracy of 52.1%. However, this is still below the necessary 52.4% to be profitable.

Polynomial Kernel (degree = 3)

Evaluation Metrics:

Accuracy: 0.528

The polynomial kernel with a degree of three performed well with an accuracy of 52.8%. This is above the necessary 52.4% to be profitable. Model accuracy did not improve using degrees past 3.

Polynomial Kernel (degree = 4)

Evaluation Metrics:

• Accuracy: 0.535

The polynomial kernel with a degree of four performed well with an accuracy of 53.5%. This is above the necessary 52.4% to be profitable. Model accuracy did not improve using degrees past 4.