# A Proper Outlier Detection For the Attached Figure

I am wondering how I could detect the red points as outliers using an algorithm (best method for this scenario) not through visualization since it is clear that they are outliers in the figure. Sometimes the red points are close to blue points but still very clear that the trend of data is broken like the attached one (80-90).

[0.00141879500243899,0.00178697795331762,0.00127892495577177,0.00143570026155573,0.00153525992083240,0.00141777922735790,0.00152684591126789,0.00148885286898430,0.00147674740998246,0.00171508938382390,0.00165398976301962,0.00142538361585036,0.00153910246593531,0.00175399308385697,0.00134698941964965,0.00172515548793911,0.00149595587129531,0.00180172725031118,0.00159037639551324,0.00187032999673476,0.00162790342840977,0.00184600681551365,0.00135449709645852,0.00176929230036940,0.00153284963291986,0.00142926162224541,0.00159764194281474,0.00163851042007631,0.00176726005837321,0.00154359218169479,0.00149738650032808,0.00152516973826902,0.00175781634013908,0.00193870678151384,0.00134159374501505,0.00166629459390643,0.00188211161641099,0.00142375632598846,0.00137379155155053,0.00183054686766502,0.00160358781806976,0.00143105342415080,0.00142689725164074,0.00133488790173896,0.00133731094914472,0.00157740531444255,0.00176267948564612,0.00158566715398123,0.00168298974941153,0.00188913573739626,0.00152608364331322,0.00161781941692366,0.00170761490900949,0.00128898986929469,0.00142269915268986,0.00153236250883160,0.00134121959772013,0.00138340739578642,0.00163948156950012,0.00158019133687677,0.00181565712864830,0.00159289331356989,0.00149272763991994,0.00150116573044485,0.00154147840865167,0.00158596715872812,0.00121074317389216,0.00156105400252430,0.00131969571973573,0.00150953288527507,0.00149105878102976,0.00147152491535477,0.00185102147226634,0.00142891794335564,0.00138431150537657,0.00151693760998362,0.00151919782691779,0.00109257864093704,0.00161380666227643,0.00190699585602371,0.000820445758286120,0.000798119153983038,0.000839861185604018,0.000861822463118421,0.000859391873730350,0.000990174686493857,0.000848612912439895,0.000916179621359871,0.00106081255138086,0.00117911733587858,0.00154535094976525,0.00144471958801459,0.00151941232953032,0.00150116342904493,0.00150758186003230,0.00174635072071058,0.00165893906442366,0.00184780915258433,0.00188991692005409,0.00150433295701926]


Thanks.

• Can you include the data used to make the chart? May 2 '19 at 21:32
• Yes. I added in the main post. May 3 '19 at 5:02

One approach would be to fit a linear regression and look at Cooks Distance to detect outliers. https://en.m.wikipedia.org/wiki/Cook%27s_distance

Some R background: http://r-statistics.co/Outlier-Treatment-With-R.html

Some Python discussion: https://stackoverflow.com/a/52322232/9524424

Here is you can find an exemplary visual comparison of binary classifiers. When I compare the plots on this website to your plot, I think, a decision tree might be a good choice in your case. I hope this is helpful.

• Thanks but I am looking for an unsupervised method May 4 '19 at 14:59