Construction of dynamic update and adaptive prediction model for user profile based on time series analysis

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

The performance of each model on different indicators

Model	F1	Accuracy/%	Precision/%	AUC	Training time/s
LR	0.894	83.62	91.94	0.731	7.18
SVM	0.925	87.4	91.07	0.803	187
DT	0.915	88.52	88.38	0.898	2.56
XGBoost	0.941	88.9	91.12	0.911	12.3
LightGBM	0.903	89.07	91.83	0.888	7.13
DCNN	0.902	89.1	92.11	0.887	1211
DF	0.909	89.9	89.08	0.833	61.12
DF+TAM+AWS	0.934	92.3	93.11	0.891	67.99

Various cascade forest classification prediction assessment

Cascade forest	F1/%	Accuracy/%	Training time/s
RF+ET+XGB+LR	93.57%	88.83%	242
RF+ET+XGB	93.32%	89.81%	93
RF+ET	93.00%	88.52%	21
RF+XGB	93.09%	88.86%	72
ET+XGB	92.01%	88.65%	79

Model training time assessment performance

Parameter	RF	ET	XGB
5	34.51	34.51	62.07
15	44.06	77.88	386.61
25	58.94	113.87	465.48
90	77.39	156.79	544.03
120	87.5	210.65	698
150	103.96	317.6	852.12
180	125.07	469.42	998.66

The performance of the training time

Parameter	RF	ET	XGB
10	21.89	32.33	68.63
25	24.4	44.71	297.96
40	23.43	30.35	485.38
55	24.01	27.02	537.97
70	23.99	27.02	557.08
85	23.02	29.29	554.83
100	21.89	38.67	492.63

The parameters of the model parameters in the cascade forest

Model	n_estimate	maxdepth
RF	15	15
ET	25	25
XGB	5	5

Model accuracy assessment performance

Parameter	RF	ET	XGB
5	0.8898	0.8852	0.8906
15	0.8905	0.8863	0.8908
25	0.8908	0.8867	0.8912
90	0.8914	0.8872	0.8917
120	0.8911	0.8867	0.8909
150	0.8911	0.8869	0.8907
180	0.8914	0.8877	0.8911

Accuracy evaluation

Parameter	RF	ET	XGB
5	0.8897	0.8635	0.8907
15	0.8898	0.8837	0.8869
25	0.889	0.8812	0.8866
90	0.8882	0.8811	0.8875
120	0.8882	0.8808	0.8873
150	0.888	0.8804	0.8872
180	0.8876	0.8801	0.8871

Experimental results

Behavior pattern	Experimental group	Control group	P
t_total	37965	38044	0.013
n_total	2365	2306	0.005
g_tests	7.18	6.84	0.236
n_rec lp/n_unrec lp	2.74	2.59	0.049
t_rec lo/t_unrec lo	2.75	0.59	0.002
n_rec lo/n_unrec lo	3.66	0.94	0.003