blog Fantasy Football Studies

Creating A Running Back Age Model To Predict Busts

After writing The Peak Age for an NFL Running Back, I turned intrigued with how age affects operating backs and how predictable the drop-off is for them. In addition, I’ve lately turn into fascinated by knowledge analysis programming languages, reminiscent of Python and R. The intersection of those two pursuits led to an try and create a predictive mannequin using operating back age and PPR scoring.

Defining a Peak Season

To win fantasy football championships, we are all trying to determine high quality or peak operating back seasons — difference-making gamers we will rely on to start out every week. Over the past two seasons, the RB10 has averaged 230 PPR fantasy points, which amounts to 14.4 fantasy points per recreation.

For the aim of this research, I included all operating back seasons since 2000 to score at the least 230 PPR fantasy points or a minimum of 200 PPR fantasy points and 14.four fantasy points per recreation. Then I charted whether the player maintained the same tempo (14.four fantasy points per recreation) in the following season. The research consists of 223 individual operating back seasons.

Word: A truthful number of seasons have been eliminated due to inconclusive outcomes. For instance, Jamaal Charles’ 2011 season was not included because within the following season he was injured in Week 2. If results for the next season were not conclusive, the season was removed.

The desk under incorporates all the seasons that have been included to create the fashions. The ‘0’ or ‘1’ underneath N + 1 signifies whether or not the participant succeeded in the following season.

Rk Player Yr Age Tm PPR PPR Per G N + 1
1 LaDainian Tomlinson* 2006 27 SDG 481.1 30.07 1
2 Marshall Faulk* 2000 27 STL 459.9 32.85 1
3 Priest Holmes 2003 30 KAN 446 27.88 1
four LaDainian Tomlinson* 2003 24 SDG 441.8 27.61 1
5 Priest Holmes 2002 29 KAN 441.7 31.55 1
6 Marshall Faulk* 2001 28 STL 422.7 30.19 1
7 Steven Jackson 2006 23 STL 415.4 25.96 1
eight David Johnson 2016 25 ARI 407.8 25.49 1
9 Edgerrin James 2000 22 IND 398.3 24.89 1
10 Chris Johnson 2009 24 TEN 395.9 24.74 1
11 Arian Foster 2010 24 HOU 393 24.56 1
12 Ahman Green 2003 26 GNB 388 24.25 1
13 Todd Gurley 2017 23 LAR 383.3 25.55 1
14 LaDainian Tomlinson* 2002 23 SDG 383.2 23.95 1
15 Jamaal Charles 2013 27 KAN 378 25.2 1
16 Ray Rice 2011 24 BAL 374.8 23.43 1
17 Shaun Alexander 2005 28 SEA 373.8 23.36 1
18 Larry Johnson 2006 27 KAN 372.9 23.31 1
19 Le’Veon Bell 2014 22 PIT 370.5 23.16 1
20 Brian Westbrook 2007 28 PHI 370.4 24.69 1
21 LaDainian Tomlinson* 2007 28 SDG 367.6 22.98 1
22 LaDainian Tomlinson* 2005 26 SDG 365.1 22.82 1
23 Ricky Williams 2002 25 MIA 363.6 22.73 1
24 Larry Johnson 2005 26 KAN 363.three 22.71 1
25 Tiki Barber 2005 30 NYG 360 22.5 1
26 DeMarco Murray 2014 26 DAL 351.1 21.94 0
27 Adrian Peterson 2012 27 MIN 347.four 21.71 1
28 Charlie Garner 2002 30 OAK 347.three 21.71 0
29 Matt Forte 2014 29 CHI 346.6 21.66 1
30 Tiki Barber 2004 29 NYG 346.6 21.66 1
31 Matt Forte 2013 28 CHI 337.three 21.08 1
32 Eddie George 2000 27 TEN 337.2 21.08 0
33 Priest Holmes 2001 28 KAN 334.9 20.93 1
34 LaDainian Tomlinson* 2004 25 SDG 334.1 22.27 1
35 Brian Westbrook 2006 27 PHI 332.6 22.17 1
36 LeSean McCoy 2013 25 PHI 330.6 20.66 0
37 LeSean McCoy 2011 23 PHI 329.4 21.96 1
38 Jamal Lewis 2003 24 BAL 329.1 20.57 0
39 Shaun Alexander 2002 25 SEA 327.5 20.47 1
40 Ray Rice 2009 22 BAL 327.1 20.44 1
41 Frank Gore 2006 23 SFO 327 20.44 1
42 Deuce McAllister 2003 25 NOR 326.7 20.42 1
43 Adrian Peterson 2009 24 MIN 325.9 20.37 1
44 Domanick Williams 2004 24 HOU 325.6 21.71 1
45 Ezekiel Elliott 2016 21 DAL 325.four 21.69 1
46 Shaun Alexander 2004 27 SEA 324.6 20.29 1
47 Tiki Barber 2002 27 NYG 324.4 20.28 1
48 Maurice Jones-Drew 2009 24 JAX 323.5 20.22 1
49 Ahman Inexperienced 2001 24 GNB 321.1 20.07 1
50 Alvin Kamara 2017 22 NOR 320.4 20.03 1
51 Ahman Green 2000 23 GNB 318.4 19.9 1
52 Le’Veon Bell 2016 24 PIT 317.4 26.45 1
53 Clinton Portis 2002 21 DEN 317.2 19.83 1
54 Curtis Martin* 2004 31 NYJ 317.2 19.83 zero
55 Devonta Freeman 2015 23 ATL 316.four 21.09 1
56 Deuce McAllister 2002 24 NOR 313 20.87 1
57 Clinton Portis 2003 22 DEN 311.5 23.96 1
58 Edgerrin James 2005 27 IND 310.3 20.69 zero
59 Curtis Martin* 2000 27 NYJ 309.9 19.37 1
60 DeAngelo Williams 2008 25 CAR 307.6 19.23 1
61 Shaun Alexander 2003 26 SEA 307 19.19 1
62 Matt Forte 2008 23 CHI 305.5 19.09 zero
63 Edgerrin James 2004 26 IND 304.1 19.01 1
64 Arian Foster 2011 25 HOU 303.1 23.32 1
65 Charlie Garner 2000 28 SFO 302.9 18.93 zero
66 Marshawn Lynch 2014 28 SEA 302.three 18.89 zero
67 Shaun Alexander 2001 24 SEA 302.1 18.88 1
68 Maurice Jones-Drew 2011 26 JAX 301 18.81 0
69 Curtis Martin* 2001 28 NYJ 299 18.69 1
70 LeSean McCoy 2016 28 BUF 298.3 19.89 1
71 Ricky Watters 2000 31 SEA 297.5 18.59 zero
72 LeSean McCoy 2010 22 PHI 297.2 19.81 1
73 Peyton Hillis 2010 24 CLE 296.9 18.56 0
74 Kareem Hunt 2017 22 KAN 295.2 18.45 1
75 LaMont Jordan 2005 27 OAK 294.8 21.06 0
76 DeMarco Murray 2016 28 TEN 293.8 18.36 zero
77 Willie Parker 2006 26 PIT 291.6 18.23 0
78 Travis Henry 2002 24 BUF 290.7 18.17 1
79 Melvin Gordon 2017 24 LAC 288.1 18.01 1
80 Tiki Barber 2000 25 NYG 287.5 17.97 1
81 Marshall Faulk* 2002 29 STL 285 20.36 1
82 Devonta Freeman 2016 24 ATL 284.1 17.76 1
83 Ray Rice 2012 25 BAL 283.1 17.69 zero
84 Michael Turner 2008 26 ATL 279 17.44 1
85 Maurice Jones-Drew 2008 23 JAX 278.9 17.43 1
86 Frank Gore 2009 26 SFO 278.6 19.9 1
87 Fred Taylor 2003 27 JAX 278.2 17.39 1
88 Mark Ingram 2017 28 NOR 278 17.38 zero
89 Brian Westbrook 2004 25 PHI 277.5 21.35 1
90 Adrian Peterson 2010 25 MIN 276.9 18.46 1
91 Clinton Portis 2007 26 WAS 276.7 17.29 1
92 Joseph Addai 2007 24 IND 276.6 18.44 zero
93 LaDainian Tomlinson* 2008 29 SDG 276.6 17.29 0
94 Mike Anderson 2000 27 DEN 276.6 17.29 0
95 Ray Rice 2010 23 BAL 276.6 17.29 1
96 Thomas Jones 2008 30 NYJ 275.9 17.24 1
97 Marshawn Lynch 2013 27 SEA 275.three 17.21 1
98 Chris Johnson 2010 25 TEN 273.9 17.12 0
99 Arian Foster 2014 28 HOU 273.5 21.04 1
100 Clinton Portis 2005 24 WAS 272.9 17.06 1
101 Steve Slaton 2008 22 HOU 272.9 17.06 1
102 Maurice Jones-Drew 2006 21 JAX 272.7 17.04 1
103 Eddie Lacy 2014 23 GNB 272.6 17.04 zero
104 LaDainian Tomlinson* 2001 22 SDG 271.3 16.96 1
105 Ahman Green 2002 25 GNB 270.3 19.31 1
106 Darren Sproles 2011 28 NOR 270.3 16.89 1
107 Brian Westbrook 2008 29 PHI 269.eight 19.27 zero
108 Marshawn Lynch 2012 26 SEA 269.6 16.85 1
109 Ricky Williams 2001 24 NOR 269.6 16.85 1
110 Darren McFadden 2010 23 OAK 269.four 20.72 1
111 Fred Taylor 2002 26 JAX 268.2 16.76 1
112 Edgerrin James 2003 25 IND 267.1 20.55 1
113 Matt Forte 2010 25 CHI 265.6 16.6 1
114 LeSean McCoy 2017 29 BUF 263.6 16.48 0
115 Reggie Bush 2006 21 NOR 262.7 16.42 1
116 Corey Dillon 2004 30 NWE 261.8 17.45 0
117 Eddie George 2002 29 TEN 261 16.31 0
118 Adrian Peterson 2015 30 MIN 260.7 16.29 1
119 Adrian Peterson 2008 23 MIN 260.5 16.28 1
120 Corey Dillon 2001 27 CIN 259.3 16.21 1
121 DeMarco Murray 2013 25 DAL 258.1 18.44 1
122 Stephen Davis 2000 26 WAS 258.1 17.21 zero
123 Lamar Smith 2000 30 MIA 258 17.2 0
124 Jamal Lewis 2002 23 BAL 257.9 16.12 1
125 C.J. Spiller 2012 25 BUF 255.three 15.96 zero
126 Duce Staley 2002 27 PHI 255 15.94 0
127 Trent Richardson 2012 22 CLE 254.7 16.98 zero
128 Adrian Peterson 2007 22 MIN 253.9 18.14 1
129 Warrick Dunn 2000 25 TAM 252.5 15.78 1
130 Alfred Morris 2012 24 WAS 252 15.75 zero
131 Chris Johnson 2008 23 TEN 250.eight 16.72 1
132 Melvin Gordon 2016 23 SDG 250.6 19.28 1
133 Jamaal Charles 2014 28 KAN 250.four 16.69 1
134 Moe Williams 2003 29 MIN 249.9 15.62 zero
135 Clinton Portis 2008 27 WAS 249.5 15.59 zero
136 Rudi Johnson 2005 26 CIN 248.eight 15.55 1
137 Tiki Barber 2003 28 NYG 247.7 15.48 1
138 Jamal Lewis 2007 28 CLE 247.2 16.48 zero
139 Justin Forsett 2014 29 BAL 246.9 15.43 0
140 Steven Jackson 2009 26 STL 246.8 16.45 1
141 Ricky Williams 2009 32 MIA 246.5 15.41 0
142 James Stewart 2000 29 DET 245.1 15.32 zero
143 Ahmad Bradshaw 2010 24 NYG 244.9 15.31 0
144 Marshawn Lynch 2011 25 SEA 244.6 16.31 1
145 Steven Jackson 2010 27 STL 243.4 15.21 1
146 Rashard Mendenhall 2010 23 PIT 243 15.19 zero
147 Eddie Lacy 2013 22 GNB 242.5 16.17 1
148 Mark Ingram 2016 27 NOR 242.2 15.14 1
149 Corey Dillon 2002 28 CIN 240.9 15.06 0
150 Chris Johnson 2013 28 TEN 240.2 15.01 0
151 Jamaal Charles 2012 26 KAN 239.5 14.97 1
152 Reggie Bush 2013 28 DET 239.2 17.09 zero
153 Frank Gore 2007 24 SFO 238.eight 15.92 1
154 Marion Barber 2007 24 DAL 238.7 14.92 1
155 Curtis Martin* 2002 29 NYJ 238.6 14.91 0
156 Thomas Jones 2009 31 NYJ 238 14.88 zero
157 Ladell Betts 2006 27 WAS 236.9 14.81 zero
158 Rudi Johnson 2004 25 CIN 236.8 14.eight 1
159 Clinton Portis 2004 23 WAS 236.6 15.77 1
160 Steven Jackson 2005 22 STL 236.6 15.77 1
161 Travis Henry 2003 25 BUF 236.4 15.76 0
162 Maurice Jones-Drew 2010 25 JAX 236.1 16.86 1
163 Ryan Mathews 2011 24 SDG 235.6 16.83 zero
164 Fred Jackson 2013 32 BUF 234.7 14.67 zero
165 Ricky Williams 2000 23 NOR 234.three 23.43 1
166 Michael Pittman 2000 25 ARI 233.8 14.61 zero
167 Carlos Hyde 2017 27 SFO 233.eight 14.61 zero
168 Michael Bennett 2002 24 MIN 233.7 14.61 0
169 Thomas Jones 2004 26 CHI 233.5 16.68 1
170 LeGarrette Blount 2016 30 NWE 232.9 14.56 zero
171 Marshawn Lynch 2008 22 BUF 232.6 15.51 0
172 Doug Martin 2015 26 TAM 232.three 14.52 0
173 Rudi Johnson 2006 27 CIN 232.three 14.52 zero
174 Tiki Barber 2001 26 NYG 232.2 16.59 1
175 Lamar Miller 2015 24 MIA 231.9 14.49 zero
176 DeAngelo Williams 2015 32 PIT 231.4 14.46 0
177 Willis McGahee 2007 26 BAL 230.8 15.39 0
178 Michael Turner 2011 29 ATL 230.eight 14.43 0
179 Warrick Dunn 2002 27 ATL 230.four 15.36 1
180 Earnest Graham 2007 27 TAM 230.2 15.35 zero
181 Leonard Fournette 2017 22 JAX 230.2 17.71 1
182 Jordan Howard 2016 22 CHI 230.1 15.34 zero
183 Ahman Inexperienced 2004 27 GNB 229.6 15.31 0
184 Domanick Williams 2003 23 HOU 229.2 16.37 1
185 Marion Barber 2008 25 DAL 229.2 15.28 0
186 Brian Westbrook 2005 26 PHI 228.3 19.03 1
187 Frank Gore 2008 25 SFO 227.9 16.28 1
188 Jamaal Charles 2009 23 KAN 227.7 15.18 1
189 Michael Pittman 2004 29 TAM 226.7 17.44 zero
190 Thomas Jones 2005 27 CHI 225.eight 15.05 zero
191 Steven Jackson 2008 25 STL 225.1 18.76 1
192 Kevin Jones 2006 24 DET 224.9 18.74 zero
193 Steven Jackson 2011 28 STL 223.eight 14.92 0
194 Chester Taylor 2006 27 MIN 223.four 14.89 0
195 Matt Forte 2011 26 CHI 222.7 18.56 1
196 Marshall Faulk* 2003 30 STL 221.8 20.16 0
197 Matt Forte 2012 27 CHI 221.4 14.76 1
198 Ahman Inexperienced 2006 29 GNB 221.2 15.8 0
199 Reggie Bush 2011 26 MIA 219.2 14.61 zero
200 Maurice Jones-Drew 2007 22 JAX 217.5 14.5 1
201 James Stewart 2002 31 DET 217.four 15.53 zero
202 Le’Veon Bell 2013 21 PIT 216.9 16.68 1
203 Mike Anderson 2005 32 DEN 216.6 14.44 0
204 Darren Sproles 2012 29 NOR 216.1 16.62 zero
205 Matt Forte 2015 30 CHI 214.7 16.52 zero
206 Deuce McAllister 2004 26 NOR 213.2 15.23 zero
207 Priest Holmes 2004 31 KAN 212.9 26.61 zero
208 Fred Jackson 2011 30 BUF 210.6 21.06 zero
209 Edgerrin James 2002 24 IND 209.3 14.95 1
210 Todd Gurley 2015 21 STL 208.4 16.03 0
211 Latavius Murray 2016 26 OAK 208.2 14.87 zero
212 Fred Taylor 2004 28 JAX 207.9 14.85 zero
213 Duce Staley 2001 26 PHI 207 15.92 1
214 DeAngelo Williams 2009 26 CAR 206.9 15.92 0
215 Reggie Bush 2007 22 NOR 206.8 17.23 1
216 Adrian Peterson 2011 26 MIN 205.9 17.16 1
217 Brandon Jacobs 2008 26 NYG 205.5 15.81 0
218 Domanick Williams 2005 25 HOU 204.3 18.57 zero
219 Mark Ingram 2015 26 NOR 203.four 16.95 1
220 Ezekiel Elliott 2017 22 DAL 203.2 20.32 1
221 Frank Gore 2010 27 SFO 202.5 18.41 zero
222 Warrick Dunn 2001 26 TAM 202.four 15.57 1
223 Steven Jackson 2007 24 STL 200.4 16.7 1

Preface

In my brief schooling in machine learning, I have discovered the next: At first, you will need to construct and train a model with a training set. Then, to research the accuracy of the mannequin, some observations have to be put aside to test the model.

Within the Kernel SVM, I neglected 20 % of the observations to test the model. Subsequently, 179 of the 223 seasons have been used as a training set. The remaining 20 %, or 44 observations, have been used to check the model.

Based mostly on the kind of knowledge being analyzed, the 2 greatest algorithms look like Kernel SVM and Naïve Bayes. We’ll begin with the Kernel SVM.

Kernel SVM Model

Let’s first understand the chart. The green factors indicate players who “succeeded” in the following season, the purple points indicate those that didn’t. The Y-axis refers to PPR factors per recreation whereas the X-axis refers to Age. The top left is younger, high-scoring players. The underside left refers to low-scoring, younger players. The highest right refers to previous, high-scoring players. The underside proper refers to previous, low-scoring players.

Word that there are far fewer points on the proper aspect of the graph — 83.eight % of the seasons befell earlier than age 29. When you embrace the age-29 seasons, it accounts for 203 of the 223 seasons — 90.6 %. It’s onerous to only make this prestigious record at 28 years previous, let alone succeeding in the following age-29 season.

This is the training model, so we’ll see how the mannequin fares in predicting the remaining 44 observations shortly. The conclusion we will draw to date is the mannequin doesn’t like the older, low-scoring seasons. The drop-off seems to take effect round age 27 for low scoring seasons. By age 29, a operating again’s play seems more likely to fall off.

Let’s examine the outcomes of Kernel SVM Check.

Out of 44 complete observations, the Kernel SVM model predicted 31 appropriately, which is an accuracy price of 70.5 %. Whereas there are some incorrect predictions on the boundary, the model does an honest job at predicting the drop-off.

Naïve Bayes Model

For the Naïve Bayes model, I used 56 observations to test the model or 25 % of the seasons. 167 observations have been used to coach the model.

The Naïve Bayes algorithm returns an identical end result – because the curve seems to comply with to the identical line. Let’s verify the outcomes of the Naïve Bayes mannequin check.

Out of 56 complete observations, the Naïve Bayes mannequin predicted 42 appropriately which is an accuracy price of 75 %. It was a bit more accurate than the Kernel SVM model.

What Does This Mean for 2019?

The model does not like older operating backs who’re coming off low scoring seasons.

Heading into last season, there were 9 gamers who matched the peak standards. Only three – Carlos Hyde, Mark Ingram, and LeSean McCoy have been over 25 years previous and thought of unlikely to take care of their tempo. All three didn’t keep the 14.four fantasy factors per recreation pace and underperformed in accordance with their common draft place.

In 2018, 13 gamers scored a minimum of 230 PPR fantasy points or no less than 200 PPR fantasy points and 14.4 fantasy factors per recreation. Saquon Barkley, Christian McCaffrey, Todd Gurley, Alvin Kamara, Ezekiel Elliott, James Conner, James White, Melvin Gordon, David Johnson, Joe Mixon, Tarik Cohen, Kareem Hunt, and Phillip Lindsay. Only three gamers have been over 24 years previous final season – Melvin Gordon (25), James White (26), and David Johnson (27).

Through the 2019 season, David Johnson will flip 28 years previous and scored simply 15.5 fantasy points per recreation in 2018. Obviously, Johnson ought to be in a a lot better state of affairs to succeed with Kliff Kingsbury coming to town, however the mannequin doesn’t like his outlook this season.

James White is 27 years previous and doesn’t obtain as many touches as the opposite players on this listing. Sharing touches with Sony Michel and Rex Burkhead will make it robust for White to proceed to supply like an RB1. One silver lining is his production with Rob Gronkowski out.

While Melvin Gordon might be 26 this yr, averaging 23 PPR fantasy points per recreation lands him within the inexperienced in response to both models. As a workhorse on an excellent offense, Gordon is at present coming off the board in the midst of the first spherical.